Synthetic Aperture Radar Image Processing and Applications

Chapter 1

Maged Marghany

1. Introduction

3

Introductory Chapter: Advanced

Ocean Current Simulation from

Satellite microwave data, such as synthetic aperture radar (SAR), have the great potential for retrieving ocean dynamic parameters, for instance, ocean surface current and ocean wave dynamic [1]. One of the attention-grabbing topics is current flow that is needed for short go back satellite cycle and high resolution. These will provide precisely data concerning current dynamic flow [2, 3]. In fact, current is very important for ship navigation, fishing, waste matter substances transport, and sediment transport [4, 5]. Respectively, optical and microwave sensors are enforced to monitor the current flows. Indeed, the ocean surface dynamic options of sea surface current are vital parameters for atmospheric-sea surface interactions. In this regard, the global climate change, marine pollution, and coastal risky are preponderantly dominated by current speed and direction [1]. The measurements of ocean current from space rely on the electromagnetic signal. Truly, associate degree of an electromagnetic signal of optical and microwave reflects from the ocean carrying records concerning one among the first discernible quantities that are the color, the beamy

Recently, the high resolution of SAR sensors such as TerraSar-X, RADARSAT-2, ALOS PALSAR, and the foremost three of the Italian satellite of COSMO-SkyMed have been commenced. Once the four satellites in the COSMO-SkyMed constellation are developed, they are conceivable functioning with a tiny resume time of a little hours [4]. Nevertheless, the initial three of the COSMO-SkyMed, ALOS PALSAR, and RADARSAT-2, satellite data are the cross-track interferometry, which do not allow determining neither coastal water flow nor coastal water level changing. In this regard, the TerraSAR-X satellite data use an along-track interferometric proficiency which simply permits the quantity of sea surface speed. Additionally, phase alterations between the coregistered pixels of an image pair are consistent to Doppler frequency shifts of the signal backscattered and according to line-of-sight velocities of the scatterers. In this view, phase alterations include influences of surface flows and of the dynamic of wave movement. Consequently, the retrieving of tidal current flow can be accurately achieved by both of TerraSAR-X and TanDEM-X. These can be depleted to regulate precisely coastal water height fluctuations. The TerraSAR-X can regulate perfectly the digital surface model (DSM), where depiction of surface-containing topographies exceeds the terrain height, for

Moreover, TanDEM-X involves dual high-resolution imaging SAR data. In this understanding, both TerraSAR-X and TanDEM-X are hovering in tandem and establishing an enormous radar interferometer with an anticipated competence of creating a comprehensive DSM through a perpendicular resolution of 2 m, exceeding

TanDEM Satellite Data

temperature, the roughness, and also the height of the ocean [2].

example, plants and constructions through precision of 2 m.

### Chapter 1

## Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

Maged Marghany

## 1. Introduction

Satellite microwave data, such as synthetic aperture radar (SAR), have the great potential for retrieving ocean dynamic parameters, for instance, ocean surface current and ocean wave dynamic [1]. One of the attention-grabbing topics is current flow that is needed for short go back satellite cycle and high resolution. These will provide precisely data concerning current dynamic flow [2, 3]. In fact, current is very important for ship navigation, fishing, waste matter substances transport, and sediment transport [4, 5]. Respectively, optical and microwave sensors are enforced to monitor the current flows. Indeed, the ocean surface dynamic options of sea surface current are vital parameters for atmospheric-sea surface interactions. In this regard, the global climate change, marine pollution, and coastal risky are preponderantly dominated by current speed and direction [1]. The measurements of ocean current from space rely on the electromagnetic signal. Truly, associate degree of an electromagnetic signal of optical and microwave reflects from the ocean carrying records concerning one among the first discernible quantities that are the color, the beamy temperature, the roughness, and also the height of the ocean [2].

Recently, the high resolution of SAR sensors such as TerraSar-X, RADARSAT-2, ALOS PALSAR, and the foremost three of the Italian satellite of COSMO-SkyMed have been commenced. Once the four satellites in the COSMO-SkyMed constellation are developed, they are conceivable functioning with a tiny resume time of a little hours [4]. Nevertheless, the initial three of the COSMO-SkyMed, ALOS PALSAR, and RADARSAT-2, satellite data are the cross-track interferometry, which do not allow determining neither coastal water flow nor coastal water level changing. In this regard, the TerraSAR-X satellite data use an along-track interferometric proficiency which simply permits the quantity of sea surface speed. Additionally, phase alterations between the coregistered pixels of an image pair are consistent to Doppler frequency shifts of the signal backscattered and according to line-of-sight velocities of the scatterers. In this view, phase alterations include influences of surface flows and of the dynamic of wave movement. Consequently, the retrieving of tidal current flow can be accurately achieved by both of TerraSAR-X and TanDEM-X. These can be depleted to regulate precisely coastal water height fluctuations. The TerraSAR-X can regulate perfectly the digital surface model (DSM), where depiction of surface-containing topographies exceeds the terrain height, for example, plants and constructions through precision of 2 m.

Moreover, TanDEM-X involves dual high-resolution imaging SAR data. In this understanding, both TerraSAR-X and TanDEM-X are hovering in tandem and establishing an enormous radar interferometer with an anticipated competence of creating a comprehensive DSM through a perpendicular resolution of 2 m, exceeding whatever obtainable currently from space [4]. Consistent with Romeiser et al. [6], with the usual helical revolution configuration, the dual satellites ensure an alongtrack gap between 0 at the northern and southern utmost locations of the orbit and approximately 550 m over the equator, restrictive of the district of convenient baselines for intersatellite interferometry above the sea surface to restricted space crews far-off the north and south. In districts of elongated along-track baselines, the data characteristic undergoes since sequential decorrelation of the signal backscattered. Nonetheless, the TanDEM-X geometry constructions acquire adjusted from period to period to enhance the cross-track interferometry performing in coastal water height fluctuations and surface stream flow attentions [7].

is collected vector height values where <sup>ς</sup> <sup>¼</sup> <sup>ς</sup>1; <sup>ς</sup>2; :……; <sup>ς</sup><sup>S</sup> ½ �T. Succeeding Baselice et al. [9] and Ferraiuolo et al. [2], a MAP algorithm height approximation can be

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

Y S

Fmc ϕ<sup>s</sup> ς<sup>s</sup> ð j Þgð Þ ς; ^σ " #

> <sup>ς</sup>s�ς<sup>k</sup> ð Þ<sup>2</sup> 2σ2

th pixel, and s is well known as

(3)

(4)

(5)

S¼1

here gð Þ: is a preceding probability density function (pdf) which is approved by means of Gaussian Markov random field (GMRF) and σ^ is the hyperparameter route which is not a preceding identified. As said by Baselice et al. [9], GMRF can be

appraised beginning from the restrained interferograms. This is realized by deliberating subbands, equivalent to diverse azimuth looks. In this regard, GMRF is determined by:

> �∑SxN <sup>S</sup>¼<sup>1</sup> <sup>∑</sup><sup>K</sup> <sup>∈</sup> NS

sk � � �

<sup>ς</sup>dMAP <sup>¼</sup> argςmaxln <sup>ð</sup>

<sup>g</sup>ð Þ¼ <sup>ς</sup>; <sup>σ</sup>^ <sup>1</sup>

where Ns is the district system of s

DOI: http://dx.doi.org/10.5772/intechopen.84644

repossess the sea surface current pattern.

throughout TanDEM-X satellite overpassed.

3.1 Satellite TanDEM-X data

3. Dataset

5

<sup>Z</sup>ð Þ <sup>σ</sup> <sup>e</sup>

[2, 9, 10]. Lastly, the regularized restoration square error is estimated via:

hyperparameters, which are illustrative of the confined physical appearance of the sea-level height h, σ is the hyperparameter vector assembling all pixel values, and Z(σ) is the detachment function [10] which is required to standardize the pdf

> <sup>ε</sup> <sup>¼</sup> k k ^<sup>ς</sup> � <sup>ς</sup> k kς 2 2

where ς is the sea-level height which is derived from Eq. (3) and the accurate height then can be estimated from Eq. (4) ð Þ^ς . Though the reform is deliberating the restricted sum of accessible data (four bands), it is virtuous to recover its feature, predominantly on the disjointedness. Formerly, inverse algorithm is executed to

Two panaches of acquaintance are required to inverse the sea surface current pattern which are: (i) TanDEM-X of SAR; and (ii) real in situ measurements

Pair of Terra-SAR satellite data is attained by the TanDEM-X satellite on May 6, 2017. The earliest date was attained at 7:27:17 am; however, the subsequent data obtained at 19:20:06 pm. Both data are in spotlight mode with X-band and HH and VV polarization, respectively. Both spotlight modes are formatted in single look complex binary data. The TanDEM-X functioning concern encompasses the synchronized maneuver of two satellites hovering in contiguous pattern. The modification restraints for the construction are: (i) the revolution arising nodes, (ii) the perspective between the perigees, (iii) the revolution peculiarities, and (iv) the phasing between the satellites. The adherence of ocean surface flow is a vigorous façade of evaluating climate variations. Space-borne SAR along-track interferometry (ATI) obligates the talent to greatly subsidize to the contemporary field.

It will recommend a great-area, global-widespread seeming surface flow quantities.

casted by:

On the word of Yoon et al. [8], the phase computation is a foremost encounter to regulate surplus precise height. This is because the calculated phase differences are assumed as a wrapped phase of the primary quantities of a scale �π to π, hence the actuality vague contained by multiples of 2π [2, 9, 10]. This technique generates phase leaps between nearby pixels. Smooth function is depleted to resolve phase leap through adding or detracting multiples of 2π. Subsequently, Ferraiuolo et al. [2] have developed the multichannel MAP height estimator as a function of a Gaussian Markov random (GMRF) to unravel the doubts of height retrieving from InSAR procedure. They initiated that the multichannel MAP height estimator has accomplished the phase gaps and tweaked the height contour as compared to predictable phase unwrapping set of rules, i.e., path-following algorithms and minimum-norm algorithms.

The foremost demonstrable of this experiment is to investigate the coastal water level and velocity changes using along-track interferometric synthetic aperture radar (ATInSAR) technique multichannel MAP height estimator.

## 2. Algorithm

The algorithm is implemented in this study, which is based on the multichannel MAP height estimator. It is depleted to retain the information of the sea surface level alterations. This algorithm is implemented from the consideration of Baselice et al. [9]. Succeeding Baselice et al. [9], the signal of interferometric phase can be articulated by the next mathematical Eq. (1) [9],

$$\phi\_{\mathfrak{m}} = \left\langle \left(\frac{4\pi}{\lambda R\_0 \sin \theta}\right) B\_{\perp \mathfrak{n}} h\_{\mathfrak{s}} + a \right\rangle\_{\mathfrak{2}\mathfrak{n}}, \mathfrak{n} = \mathbf{1}, \mathbf{2}, \dots, \mathbf{N}; \mathfrak{s} = \mathbf{1}, \mathbf{2}, \dots, \mathbf{S} \tag{1}$$

where s is the pixel locus in the TanDEM-X data, n is the deliberated interferogram band, λ is the TanDEM-X wavelength, R<sup>0</sup> is the stretch between the epicenter of the sight and the controlling antenna, and B⊥<sup>n</sup> is the orthogonal baseline. Furthermore, hs is the height rate in meter, α is the phase decorrelation noise, and an incident angle is presented by θ. Moreover, h i: <sup>2</sup><sup>π</sup> signifies the "modulo-2π." Let us assume that N is autonomous interferogram bands; thenceforth, the obstruction contains the retrieving of the sea-level height rates hs, which is being from the S � N as a function of the expected wrapped phase ϕsn. Succeeding Ferraiuolo et al. [2], the obstruction of demonstrating height can be elucidated by means of a MAP height approximation technique. In this understanding, the multichannel probability function Fmc is formulated as:

$$F\_{mc}(\phi\_s|\mathfrak{c}\_s) = \prod\_{n=1}^{N} f((\phi\_{sn}|\mathfrak{c}\_s) \tag{2}$$

here F ϕsn ς<sup>s</sup> ð j Þ is the likelihood function of the signal channel, ϕ<sup>s</sup> is calculated as wrapped phase data which is denoted as the pixel s, ϕ<sup>s</sup> ¼ ϕ<sup>s</sup>1; ϕ<sup>s</sup><sup>2</sup> ½ � ; :……; ϕsN <sup>T</sup>, and ς<sup>s</sup> Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data DOI: http://dx.doi.org/10.5772/intechopen.84644

is collected vector height values where <sup>ς</sup> <sup>¼</sup> <sup>ς</sup>1; <sup>ς</sup>2; :……; <sup>ς</sup><sup>S</sup> ½ �T. Succeeding Baselice et al. [9] and Ferraiuolo et al. [2], a MAP algorithm height approximation can be casted by:

$$\widehat{\varphi\_{MAP}} = \arg\max\_{\boldsymbol{\xi}} \ln \left[ \left( \prod\_{\mathcal{S}=1}^{\mathcal{S}} F\_{mc}(\phi\_{\boldsymbol{\varepsilon}}|\boldsymbol{\xi}\_{\boldsymbol{\varepsilon}}) \mathbf{g}(\boldsymbol{\varepsilon}; \boldsymbol{\hat{\sigma}}) \right) \right] \tag{3}$$

here gð Þ: is a preceding probability density function (pdf) which is approved by means of Gaussian Markov random field (GMRF) and σ^ is the hyperparameter route which is not a preceding identified. As said by Baselice et al. [9], GMRF can be appraised beginning from the restrained interferograms. This is realized by deliberating subbands, equivalent to diverse azimuth looks. In this regard, GMRF is determined by:

$$\log(\boldsymbol{\varepsilon}, \hat{\boldsymbol{\sigma}}) = \frac{1}{Z(\boldsymbol{\sigma})} \boldsymbol{\varepsilon}^{\left(-\sum\_{S=1}^{\text{Sr}N} \sum\_{K \in N\_S} \left[\frac{\left(\boldsymbol{\varepsilon} - \boldsymbol{\varepsilon}\_k\right)^2}{2\sigma\_{\text{sk}}^2}\right]\right)}{} \tag{4}$$

where Ns is the district system of s th pixel, and s is well known as hyperparameters, which are illustrative of the confined physical appearance of the sea-level height h, σ is the hyperparameter vector assembling all pixel values, and Z(σ) is the detachment function [10] which is required to standardize the pdf [2, 9, 10]. Lastly, the regularized restoration square error is estimated via:

$$\varepsilon = \frac{\left\|\hat{\boldsymbol{\varepsilon}} - \boldsymbol{\varepsilon}\right\|^2}{\left\|\boldsymbol{\varepsilon}\right\|^2} \tag{5}$$

where ς is the sea-level height which is derived from Eq. (3) and the accurate height then can be estimated from Eq. (4) ð Þ^ς . Though the reform is deliberating the restricted sum of accessible data (four bands), it is virtuous to recover its feature, predominantly on the disjointedness. Formerly, inverse algorithm is executed to repossess the sea surface current pattern.

#### 3. Dataset

whatever obtainable currently from space [4]. Consistent with Romeiser et al. [6], with the usual helical revolution configuration, the dual satellites ensure an alongtrack gap between 0 at the northern and southern utmost locations of the orbit and approximately 550 m over the equator, restrictive of the district of convenient baselines for intersatellite interferometry above the sea surface to restricted space crews far-off the north and south. In districts of elongated along-track baselines, the data characteristic undergoes since sequential decorrelation of the signal backscattered. Nonetheless, the TanDEM-X geometry constructions acquire adjusted from period to period to enhance the cross-track interferometry performing in coastal water height

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

On the word of Yoon et al. [8], the phase computation is a foremost encounter to regulate surplus precise height. This is because the calculated phase differences are assumed as a wrapped phase of the primary quantities of a scale �π to π, hence the actuality vague contained by multiples of 2π [2, 9, 10]. This technique generates phase leaps between nearby pixels. Smooth function is depleted to resolve phase leap through adding or detracting multiples of 2π. Subsequently, Ferraiuolo et al. [2] have developed the multichannel MAP height estimator as a function of a Gaussian Markov random (GMRF) to unravel the doubts of height retrieving from InSAR procedure. They initiated that the multichannel MAP height estimator has accomplished the phase gaps and tweaked the height contour as compared to predictable phase unwrapping set of

The foremost demonstrable of this experiment is to investigate the coastal water

The algorithm is implemented in this study, which is based on the multichannel MAP height estimator. It is depleted to retain the information of the sea surface level alterations. This algorithm is implemented from the consideration of Baselice et al. [9]. Succeeding Baselice et al. [9], the signal of interferometric phase can be

2π

where s is the pixel locus in the TanDEM-X data, n is the deliberated interferogram band, λ is the TanDEM-X wavelength, R<sup>0</sup> is the stretch between the epicenter of the sight and the controlling antenna, and B⊥<sup>n</sup> is the orthogonal baseline. Furthermore, hs is the height rate in meter, α is the phase decorrelation noise, and an incident angle is presented by θ. Moreover, h i: <sup>2</sup><sup>π</sup> signifies the "modulo-2π." Let us assume that N is autonomous interferogram bands; thenceforth, the obstruction contains the retrieving of the sea-level height rates hs, which is being from the S � N as a function of the expected wrapped phase ϕsn. Succeeding Ferraiuolo et al. [2], the obstruction of demonstrating height can be elucidated by means of a MAP height approximation technique. In this understanding, the multichannel probabil-

, n ¼ 1, 2, …, N; s ¼ 1, 2, …, S (1)

f ϕsn ς<sup>s</sup> ðð j Þ (2)

<sup>T</sup>, and ς<sup>s</sup>

level and velocity changes using along-track interferometric synthetic aperture

rules, i.e., path-following algorithms and minimum-norm algorithms.

radar (ATInSAR) technique multichannel MAP height estimator.

B⊥nhs þ α

Fmc <sup>ϕ</sup><sup>s</sup> <sup>ς</sup><sup>s</sup> ð Þ¼ <sup>j</sup> <sup>Y</sup>

wrapped phase data which is denoted as the pixel s, ϕ<sup>s</sup> ¼ ϕ<sup>s</sup>1; ϕ<sup>s</sup><sup>2</sup> ½ � ; :……; ϕsN

N

n¼1

here F ϕsn ς<sup>s</sup> ð j Þ is the likelihood function of the signal channel, ϕ<sup>s</sup> is calculated as

articulated by the next mathematical Eq. (1) [9],

λR<sup>0</sup> sin θ � �

� �

<sup>ϕ</sup>sn <sup>¼</sup> <sup>4</sup><sup>π</sup>

ity function Fmc is formulated as:

4

2. Algorithm

fluctuations and surface stream flow attentions [7].

Two panaches of acquaintance are required to inverse the sea surface current pattern which are: (i) TanDEM-X of SAR; and (ii) real in situ measurements throughout TanDEM-X satellite overpassed.

#### 3.1 Satellite TanDEM-X data

Pair of Terra-SAR satellite data is attained by the TanDEM-X satellite on May 6, 2017. The earliest date was attained at 7:27:17 am; however, the subsequent data obtained at 19:20:06 pm. Both data are in spotlight mode with X-band and HH and VV polarization, respectively. Both spotlight modes are formatted in single look complex binary data. The TanDEM-X functioning concern encompasses the synchronized maneuver of two satellites hovering in contiguous pattern. The modification restraints for the construction are: (i) the revolution arising nodes, (ii) the perspective between the perigees, (iii) the revolution peculiarities, and (iv) the phasing between the satellites. The adherence of ocean surface flow is a vigorous façade of evaluating climate variations. Space-borne SAR along-track interferometry (ATI) obligates the talent to greatly subsidize to the contemporary field. It will recommend a great-area, global-widespread seeming surface flow quantities.

## Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

The difficulties of representing comparatively low-slung speeds are regular resolute by the developments of SAR satellites that produce satisfactorily considerate ATI quantities [7].

was collected by the Aquadopp® 2 MHz current meter factory-made by Nortek AS, Scandinavian country. The device could be a standalone composition, manipulation of the Doppler-established frequency equipment to gauge the surface current flows at the positioning of a fixed geographical location on the sea surface. The equipment is envisioned basically with memory and internal battery pack somewhere it may be

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

Along the coastal water of Teluk Kemang, Port Dickson, Malaysia, the current meter instrument of Aquadopp® 2 MHz current meter was arrayed on May 6, 2017 (Figure 2). Two periods of data collection were carried out: (i) at 6:15 am to 8:15 am and (ii) at 6:15 pm to 8:15 pm. For both phases, therefore, the surface flows were

intended to tape and collect information within for self-positioning [3].

The TanDEM-X satellite data with the spotlight of VV polarization are implemented to retrieve the sea surface flow rates (Figure 3). The retrieving sea level and sea surface flow variations are constrained to range direction. In fact, the sea surface current is only sensed along the range, while the wave spectra information is a function of SAR azimuth direction. The retrieving sea surface flows are delivered inshore zone of the coastal water of the Teluk Kemang, Port Dickson as

Therefore, the Doppler shift frequency of the ATI indicates fluctuations of sea surface flow. The inshore water has a weak flow along 5 km of the coastal water. This is indicated by the lower rate value of 0.1 m/s. In this regard, the lowest spectral peak of the Doppler frequency shift is 0.04 which is corresponding to the frequency shift value of 200 Hz (Figure 4). In this view, the weak inshore water flow could be attributed to the impact of the low tide of 0.3 m as noticed along the

The interferogram phase is ranged between 0.7° and +0.7° (Figure 5) derived by the multichannel MAP height estimator. Obviously, the same pattern is visible.

deliberated for intermissions of 2 h.

DOI: http://dx.doi.org/10.5772/intechopen.84644

part of the Malacca Straits.

Figure 3.

7

4. Current pattern from TanDEM-X data

coastal water of the Teluk Kemang, Port Dickson.

TanDEM-X SAR data (a) first mission and (b) second mission with VV polarization.

In this revision, the multichannel MAP height estimator relies on the TanDEMX facts. Both TerraSAR-X and TanDEM-X satellites transmit identical SAR devices functioning at 9.65 GHz frequency (X-band). All over approximately dedicated maneuvers, both satellites are positioned acquaintance exceptionally in an actual singular track conformation through a fleeting along path reference line delivering a possibility for sea surface flow quantities. The TanDEM-X data exploited in this investigation were bistatic (TS-X active/TD-X passive) channel with VV polarization and in stripmap (SM) [6, 7].

## 3.2 In situ ocean current measurement

Succeeding Marghany [3], the device of Aquadopp® 2 MHz current meter was used to acquire the physical information of sea surface flows, for instance, speed and direction (Figure 1). In this view, the surface flow information achievement

Figure 1. Deployment of Aquadop 2 MHz current meter in the coastal water.

Figure 2. In situ measurements geographical location.

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data DOI: http://dx.doi.org/10.5772/intechopen.84644

was collected by the Aquadopp® 2 MHz current meter factory-made by Nortek AS, Scandinavian country. The device could be a standalone composition, manipulation of the Doppler-established frequency equipment to gauge the surface current flows at the positioning of a fixed geographical location on the sea surface. The equipment is envisioned basically with memory and internal battery pack somewhere it may be intended to tape and collect information within for self-positioning [3].

Along the coastal water of Teluk Kemang, Port Dickson, Malaysia, the current meter instrument of Aquadopp® 2 MHz current meter was arrayed on May 6, 2017 (Figure 2). Two periods of data collection were carried out: (i) at 6:15 am to 8:15 am and (ii) at 6:15 pm to 8:15 pm. For both phases, therefore, the surface flows were deliberated for intermissions of 2 h.

### 4. Current pattern from TanDEM-X data

The TanDEM-X satellite data with the spotlight of VV polarization are implemented to retrieve the sea surface flow rates (Figure 3). The retrieving sea level and sea surface flow variations are constrained to range direction. In fact, the sea surface current is only sensed along the range, while the wave spectra information is a function of SAR azimuth direction. The retrieving sea surface flows are delivered inshore zone of the coastal water of the Teluk Kemang, Port Dickson as part of the Malacca Straits.

Therefore, the Doppler shift frequency of the ATI indicates fluctuations of sea surface flow. The inshore water has a weak flow along 5 km of the coastal water. This is indicated by the lower rate value of 0.1 m/s. In this regard, the lowest spectral peak of the Doppler frequency shift is 0.04 which is corresponding to the frequency shift value of 200 Hz (Figure 4). In this view, the weak inshore water flow could be attributed to the impact of the low tide of 0.3 m as noticed along the coastal water of the Teluk Kemang, Port Dickson.

The interferogram phase is ranged between 0.7° and +0.7° (Figure 5) derived by the multichannel MAP height estimator. Obviously, the same pattern is visible.

Figure 3. TanDEM-X SAR data (a) first mission and (b) second mission with VV polarization.

The difficulties of representing comparatively low-slung speeds are regular resolute by the developments of SAR satellites that produce satisfactorily considerate ATI

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

In this revision, the multichannel MAP height estimator relies on the TanDEMX facts. Both TerraSAR-X and TanDEM-X satellites transmit identical SAR devices functioning at 9.65 GHz frequency (X-band). All over approximately dedicated maneuvers, both satellites are positioned acquaintance exceptionally in an actual singular track conformation through a fleeting along path reference line delivering a possibility for sea surface flow quantities. The TanDEM-X data exploited in this investigation were bistatic (TS-X active/TD-X passive) channel with VV polariza-

Succeeding Marghany [3], the device of Aquadopp® 2 MHz current meter was used to acquire the physical information of sea surface flows, for instance, speed and direction (Figure 1). In this view, the surface flow information achievement

quantities [7].

Figure 1.

Figure 2.

6

In situ measurements geographical location.

tion and in stripmap (SM) [6, 7].

3.2 In situ ocean current measurement

Deployment of Aquadop 2 MHz current meter in the coastal water.

Figure 4. TanDEM-X data Doppler spectra intensity.

Figure 6 exhibits noteworthy correspondences between the consequence of sea

As said by Romeiser et al. [6, 7], the signatures of the Doppler frequency shift are clearly responsive to sea surface flow than to wind modifications. Similarly, a modification of the Doppler frequency shifts has a tiny effect on the TanDEM-X backscatter intensity as compared to relaxation rate. Obtaining phase by using multichannel MAP height estimator algorithm will allow us to characterize the water sea-level fluctuations. Three-dimensional reconstruction of water-level

surface flow speeds, which are created from TanDEM-X satellite data, and the consequence delivered in the in situ quantity. Figure 7 demonstrates how the correlation coefficient alteration as direct correlation between the two different parameters is modified. Indeed, the investigation of the correlation between different measured parameters can assist to develop accurate model. Obviously, there is a worthy correlation between the retrieved sea surface flow and real in situ measured flow with r2 of 0.76. Conversely, this correlation is not faultless, but it appears to have a confident, direct association, and resembles to what one would guess when bearing in mind both sea surface flow simulation from satellite data and one is

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

DOI: http://dx.doi.org/10.5772/intechopen.84644

measured in situ and then follow the hypothesis of normality.

Sea surface current and sea-level variations retrieved using an ATI MAP algorithm.

Figure 6.

9

Figure 5. Pattern of interferogram phase.

This pattern signature represents current feature variations along the coastal waters. Conversely, this interferogram phase is dominated by noises.

The inverters of the interferogram phase can be used to compute the ATI Doppler sea surface current. The ATI Doppler shows a clear current pattern movement along the coastal waters with minimum and maximum speed of 0.1 and 0.2 m/s, correspondingly (Figure 6). In fact, the interferometric combination of the two images reveals phase alterations that are comparable to the backscatter variations of the Doppler frequency shift [7]. This rapidity is conforming to sea-level differences of 0.4 m (Figure 6).

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data DOI: http://dx.doi.org/10.5772/intechopen.84644

Figure 6.

Sea surface current and sea-level variations retrieved using an ATI MAP algorithm.

Figure 6 exhibits noteworthy correspondences between the consequence of sea surface flow speeds, which are created from TanDEM-X satellite data, and the consequence delivered in the in situ quantity. Figure 7 demonstrates how the correlation coefficient alteration as direct correlation between the two different parameters is modified. Indeed, the investigation of the correlation between different measured parameters can assist to develop accurate model. Obviously, there is a worthy correlation between the retrieved sea surface flow and real in situ measured flow with r2 of 0.76. Conversely, this correlation is not faultless, but it appears to have a confident, direct association, and resembles to what one would guess when bearing in mind both sea surface flow simulation from satellite data and one is measured in situ and then follow the hypothesis of normality.

As said by Romeiser et al. [6, 7], the signatures of the Doppler frequency shift are clearly responsive to sea surface flow than to wind modifications. Similarly, a modification of the Doppler frequency shifts has a tiny effect on the TanDEM-X backscatter intensity as compared to relaxation rate. Obtaining phase by using multichannel MAP height estimator algorithm will allow us to characterize the water sea-level fluctuations. Three-dimensional reconstruction of water-level

This pattern signature represents current feature variations along the coastal

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

The inverters of the interferogram phase can be used to compute the ATI Doppler sea surface current. The ATI Doppler shows a clear current pattern movement along the coastal waters with minimum and maximum speed of 0.1 and 0.2 m/s, correspondingly (Figure 6). In fact, the interferometric combination of the two images reveals phase alterations that are comparable to the backscatter variations of the Doppler frequency shift [7]. This rapidity is conforming to sea-level

waters. Conversely, this interferogram phase is dominated by noises.

differences of 0.4 m (Figure 6).

Figure 5.

8

Figure 4.

TanDEM-X data Doppler spectra intensity.

Pattern of interferogram phase.

Figure 7. Validation of MAP algorithm with in situ measurement.

changes from the ATInSAR technique by using the algorithm of multichannel MAP height approximation can aid to regulate the vertical shift of sea-level changes. Moreover, the multichannel MAP height approximation has achieved the difficulty of the phase unwrapping discontinuities and amended the vertical displacement synopsis as rivaled to conservative algorithm of phase unwrapping, for instance, (i) minimum-norm algorithm and (ii) path-following algorithm [9]. Lastly, TanDEM-X satellite data are comprehended as the prospective radar device for observing the dynamic fluctuation of ocean surface. Sea surface flow is considered as one of a consideration—impressing issue which is required a short visit cycle and extraordinary resolution. In this understanding, these can afford specific facts in relation to sea surface dynamic flow [2, 3, 5, 9–16].

## 5. Conclusion

This work has revealed a method for regaining sea surface flow using such highresolution satellite data of TanDEM SAR-X. Along-track interferometry (ATI) technique is implemented to retrieve sea surface current movement. To this end, multichannel MAP height estimator algorithm is said to model sea-level variation. Then, the inverse algorithm is used which is based on the Doppler frequency model to retrieve sea surface current. The results reveal that the sea surface flow pattern is dominated by low velocity of less than 0.3 m/s which corresponds to lower sea-level variation of 0.4 m. The study confirms that multichannel MAP height estimator algorithm is proficient to regain the sea surface flow rate from ATI TanDEM-X with an extraordinary precision of 0.09 m/s. In conclusion, the approximation algorithm of multichannel MAP height conceivably can be a tremendous practice for repossessing sea surface flow pattern and sea-level fluctuations from ATI TanDEM-X satellite data.

Author details

Maged Marghany

Malaysia

11

Faculty Geospatial and Real Estate, Geomatika University College, Kuala Lumpur,

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

DOI: http://dx.doi.org/10.5772/intechopen.84644

© 2019 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium,

\*Address all correspondence to: magedupm@hotmail.com

provided the original work is properly cited.

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data DOI: http://dx.doi.org/10.5772/intechopen.84644

## Author details

changes from the ATInSAR technique by using the algorithm of multichannel MAP height approximation can aid to regulate the vertical shift of sea-level changes. Moreover, the multichannel MAP height approximation has achieved the difficulty of the phase unwrapping discontinuities and amended the vertical displacement synopsis as rivaled to conservative algorithm of phase unwrapping, for instance, (i) minimum-norm algorithm and (ii) path-following algorithm [9]. Lastly, TanDEM-X satellite data are comprehended as the prospective radar device for observing the dynamic fluctuation of ocean surface. Sea surface flow is considered as one of a consideration—impressing issue which is required a short visit cycle and extraordinary resolution. In this understanding, these can afford specific facts in relation to

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

This work has revealed a method for regaining sea surface flow using such high-

resolution satellite data of TanDEM SAR-X. Along-track interferometry (ATI) technique is implemented to retrieve sea surface current movement. To this end, multichannel MAP height estimator algorithm is said to model sea-level variation. Then, the inverse algorithm is used which is based on the Doppler frequency model to retrieve sea surface current. The results reveal that the sea surface flow pattern is dominated by low velocity of less than 0.3 m/s which corresponds to lower sea-level variation of 0.4 m. The study confirms that multichannel MAP height estimator algorithm is proficient to regain the sea surface flow rate from ATI TanDEM-X with an extraordinary precision of 0.09 m/s. In conclusion, the approximation algorithm of multichannel MAP height conceivably can be a tremendous practice for repossessing sea surface flow pattern and sea-level fluctuations from ATI TanDEM-X

sea surface dynamic flow [2, 3, 5, 9–16].

Validation of MAP algorithm with in situ measurement.

5. Conclusion

Figure 7.

satellite data.

10

Maged Marghany Faculty Geospatial and Real Estate, Geomatika University College, Kuala Lumpur, Malaysia

\*Address all correspondence to: magedupm@hotmail.com

© 2019 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

## References

[1] Lu Z, Kim J-W, Lee H, Shum C, Duan J, Ibaraki M, et al. Helmand river hydrologic studies using ALOS PALSAR InSAR and ENVISAT altimetry. Marine Geodesy. 2009;32(3):320-333

[2] Ferraiuolo G, Meglio F, Pascazio V, Schirinzi G. DEM reconstruction accuracy in multichannel SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2009; 47(1):191-201

[3] Marghany M. Developing robust model for retrieving sea surface current from RADARSAT-1 SAR satellite data. International Journal of Physical Sciences. 2011;6(29):6630-6637

[4] Mason DC, Speck R, Devereux B, Schumann GJ-P, Neal JC, Bates PD. Flood detection in urban areas using TerraSAR-X. IEEE Transactions on Geoscience and Remote Sensing. 2010; 48(2):882-894

[5] Marghany M. Three-dimensional visualisation of coastal geomorphology using fuzzy B-spline of dinsar technique. International Journal of Physical Sciences. 2011;6(30):6967-6971

[6] Romeiser R, Runge H. Theoretical evaluation of several possible alongtrack InSAR modes of TerraSAR-X for ocean current measurements. IEEE Transactions on Geoscience and Remote Sensing. 2007;45(1):21-35

[7] Romeiser R, Runge H, Suchandt S, Kahle R, Rossi C, Bell PS. Quality assessment of surface current fields from TerraSAR-X and TanDEM-X along-track interferometry and Doppler centroid analysis. IEEE Transactions on Geoscience and Remote Sensing. 2014; 52(5):2759-2772

[8] Yoon G-W, Kim S-W, Lee Y-W, Won J-S. Measurement of the water level in reservoirs from TerraSAR-X SAR interferometry and amplitude images. Remote Sensing Letters. 2013; 4(5):446-454

[16] Marghany M. Simulation sea surface current from RADARSAT-2 SAR data using Hopfield neural network. In: Synthetic Aperture Radar (APSAR), 2015 IEEE 5th Asia-Pacific Conference on Synthetic Aperture Radar (APSAR). New York: Institute of Electrical and Electronics Engineers (IEEE); 2015.

DOI: http://dx.doi.org/10.5772/intechopen.84644

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data

pp. 805-808

13

[9] Baselice F, Ferraioli G, Pascazio V. DEM reconstruction in layover areas from SAR and auxiliary input data. IEEE Geoscience and Remote Sensing Letters. 2009;6(2):253-257

[10] Ferraiuolo G, Pascazio V, Schirinzi G. Maximum a posteriori estimation of height profiles in InSAR imaging. IEEE Geoscience and Remote Sensing Letters. 2004;1(2):66-70

[11] Marghany M. Three-dimensional coastal geomorphology deformation modelling using differential synthetic aperture interferometry. Zeitschrift fur Naturforschung A: Journal of Physical Sciences. 2012;67(6):419

[12] Marghany M. DEM reconstruction of coastal geomorphology from DINSAR. In: Murgante B et al., editors. Lecture Notes in Computer Science (ICCSA 2012). Part III, LNCS 7335. New York City: Springer; 2012. pp. 435-446

[13] Marghany M. DInSAR technique for three-dimensional coastal spit simulation from radarsat-1 fine mode data. Acta Geophysica. 2013;61(2):478-493

[14] Marghany M. Simulation of Tsunami Impact on Sea Surface Salinity along Banda Aceh Coastal Waters, Indonesia. In: Marghany M, editors. Advanced Geoscience Remote Sensing. Croatia: Intech; 2014. pp. 229-251

[15] Marghany M. Hybrid genetic algorithm of interferometric synthetic aperture radar for three-dimensional coastal deformation. In: Hybrid Genetic Algorithm of Interferometric Synthetic Aperture Radar for Three-Dimensional Coastal Deformation. 2014. pp. 116-131

Introductory Chapter: Advanced Ocean Current Simulation from TanDEM Satellite Data DOI: http://dx.doi.org/10.5772/intechopen.84644

[16] Marghany M. Simulation sea surface current from RADARSAT-2 SAR data using Hopfield neural network. In: Synthetic Aperture Radar (APSAR), 2015 IEEE 5th Asia-Pacific Conference on Synthetic Aperture Radar (APSAR). New York: Institute of Electrical and Electronics Engineers (IEEE); 2015. pp. 805-808

References

47(1):191-201

48(2):882-894

[1] Lu Z, Kim J-W, Lee H, Shum C, Duan J, Ibaraki M, et al. Helmand river hydrologic studies using ALOS PALSAR InSAR and ENVISAT altimetry. Marine

level in reservoirs from TerraSAR-X SAR interferometry and amplitude images. Remote Sensing Letters. 2013;

[9] Baselice F, Ferraioli G, Pascazio V. DEM reconstruction in layover areas from SAR and auxiliary input data. IEEE Geoscience and Remote Sensing Letters.

[10] Ferraiuolo G, Pascazio V, Schirinzi G. Maximum a posteriori estimation of height profiles in InSAR imaging. IEEE Geoscience and Remote Sensing Letters.

[11] Marghany M. Three-dimensional coastal geomorphology deformation modelling using differential synthetic aperture interferometry. Zeitschrift fur Naturforschung A: Journal of Physical

[12] Marghany M. DEM reconstruction of coastal geomorphology from

DINSAR. In: Murgante B et al., editors. Lecture Notes in Computer Science (ICCSA 2012). Part III, LNCS 7335. New York City: Springer; 2012. pp. 435-446

[13] Marghany M. DInSAR technique for three-dimensional coastal spit simulation from radarsat-1 fine mode data. Acta Geophysica. 2013;61(2):478-493

[14] Marghany M. Simulation of Tsunami Impact on Sea Surface Salinity along Banda Aceh Coastal Waters, Indonesia. In: Marghany M, editors. Advanced Geoscience Remote Sensing. Croatia:

Intech; 2014. pp. 229-251

[15] Marghany M. Hybrid genetic algorithm of interferometric synthetic aperture radar for three-dimensional coastal deformation. In: Hybrid Genetic Algorithm of Interferometric Synthetic Aperture Radar for Three-Dimensional Coastal Deformation. 2014. pp. 116-131

4(5):446-454

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

2009;6(2):253-257

2004;1(2):66-70

Sciences. 2012;67(6):419

[2] Ferraiuolo G, Meglio F, Pascazio V, Schirinzi G. DEM reconstruction accuracy in multichannel SAR

interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2009;

[3] Marghany M. Developing robust model for retrieving sea surface current from RADARSAT-1 SAR satellite data. International Journal of Physical Sciences. 2011;6(29):6630-6637

[4] Mason DC, Speck R, Devereux B, Schumann GJ-P, Neal JC, Bates PD. Flood detection in urban areas using TerraSAR-X. IEEE Transactions on Geoscience and Remote Sensing. 2010;

[5] Marghany M. Three-dimensional visualisation of coastal geomorphology

[6] Romeiser R, Runge H. Theoretical evaluation of several possible alongtrack InSAR modes of TerraSAR-X for ocean current measurements. IEEE Transactions on Geoscience and Remote

[7] Romeiser R, Runge H, Suchandt S, Kahle R, Rossi C, Bell PS. Quality assessment of surface current fields from TerraSAR-X and TanDEM-X along-track interferometry and Doppler centroid analysis. IEEE Transactions on Geoscience and Remote Sensing. 2014;

[8] Yoon G-W, Kim S-W, Lee Y-W, Won J-S. Measurement of the water

using fuzzy B-spline of dinsar technique. International Journal of Physical Sciences. 2011;6(30):6967-6971

Sensing. 2007;45(1):21-35

52(5):2759-2772

12

Geodesy. 2009;32(3):320-333

Chapter 2

Abstract

On Feature-Based SAR Image

and Retrieval Algorithm

scheme is validated on both InSAR and MiniSAR image pairs.

synthetic aperture radar (SAR)

1. Introduction

15

Keywords: extended fast least trimmed squares (EF-LTS), feature-based image

Synthetic aperture radar (SAR) as an irreplaceable remote sensing technique has been used for earth observation and environment monitoring for a long time due to its all-weather and all-day operational capability. A large number of airborne and spaceborne SAR sensors have been deployed recently. Nevertheless, the difference in sensors and imaging geometries will always introduce a geometrical warp between images which should be compensated before any joint application of multiple SAR images for accurate apperception and understanding of target and scene. Image registration is just dedicated to retrieve the warp function to align the same

A lot of SAR image registration techniques have been developed hitherto. In this

registration, parameter estimation, speeded up robust feature (SURF),

pixel position in each SAR image to the same target in the global system.

chapter, we focus on the algorithms that conduct registration based on image features, such as contour, region, line, and point. Contour, region, and line as well

Dong Li, Yunhua Zhang and Xiaojin Shi

Registration: Appropriate Feature

An investigation on the appropriate feature and parameter retrieval algorithm is conducted for feature-based registration of synthetic aperture radar (SAR) images. The commonly used features such as tie points, Harris corner, SIFT, and SURF are comprehensively evaluated. SURF is shown to outperform others on criteria such as the geometrical invariance of feature and descriptor, the extraction and matching speed, the localization accuracy, as well as the robustness to decorrelation and speckling. The processing result reveals that SURF has nice flexibility to SAR speckles for the potential relationship between Fast-Hessian detector and refined Lee filter. Moreover, the use of Fast-Hessian to oversampled images with unaltered sampling step helps to improve the registration accuracy to subpixel (i.e., <1 pixel). As for parameter retrieval, the widely used random sample consensus (RANSAC) is inappropriate because it may trap into local occlusion and result in uncertain estimation. An extended fast least trimmed squares (EF-LTS) is proposed, which behaves stable and averagely better than RANSAC. Fitting SURF features with EF-LTS is hence suggested for SAR image registration. The nice performance of this

## Chapter 2

## On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

Dong Li, Yunhua Zhang and Xiaojin Shi

## Abstract

An investigation on the appropriate feature and parameter retrieval algorithm is conducted for feature-based registration of synthetic aperture radar (SAR) images. The commonly used features such as tie points, Harris corner, SIFT, and SURF are comprehensively evaluated. SURF is shown to outperform others on criteria such as the geometrical invariance of feature and descriptor, the extraction and matching speed, the localization accuracy, as well as the robustness to decorrelation and speckling. The processing result reveals that SURF has nice flexibility to SAR speckles for the potential relationship between Fast-Hessian detector and refined Lee filter. Moreover, the use of Fast-Hessian to oversampled images with unaltered sampling step helps to improve the registration accuracy to subpixel (i.e., <1 pixel). As for parameter retrieval, the widely used random sample consensus (RANSAC) is inappropriate because it may trap into local occlusion and result in uncertain estimation. An extended fast least trimmed squares (EF-LTS) is proposed, which behaves stable and averagely better than RANSAC. Fitting SURF features with EF-LTS is hence suggested for SAR image registration. The nice performance of this scheme is validated on both InSAR and MiniSAR image pairs.

Keywords: extended fast least trimmed squares (EF-LTS), feature-based image registration, parameter estimation, speeded up robust feature (SURF), synthetic aperture radar (SAR)

## 1. Introduction

Synthetic aperture radar (SAR) as an irreplaceable remote sensing technique has been used for earth observation and environment monitoring for a long time due to its all-weather and all-day operational capability. A large number of airborne and spaceborne SAR sensors have been deployed recently. Nevertheless, the difference in sensors and imaging geometries will always introduce a geometrical warp between images which should be compensated before any joint application of multiple SAR images for accurate apperception and understanding of target and scene. Image registration is just dedicated to retrieve the warp function to align the same pixel position in each SAR image to the same target in the global system.

A lot of SAR image registration techniques have been developed hitherto. In this chapter, we focus on the algorithms that conduct registration based on image features, such as contour, region, line, and point. Contour, region, and line as well

or minimizing the difference. The warp function can then be retrieved by fitting the obtained correspondences. For correspondences without any mismatches, the retrieval can be easily conducted by fitting them with the least squares (LS). However, for the general registration cases, the initial correspondences often contain mismatches. Therefore, the robust retrieval algorithms which are insensitive to outliers are needed. In many existing literatures on feature-based SAR image registration [15, 16, 26, 27], the random sample consensus (RANSAC) [30] has been widely used and recommended for warp function retrieval. RANSAC conducts the estimation by randomly sampling a minimal sampling set (MSS) to achieve an estimation of the warping, and the entire datasets are then checked on the estimation for a consensus set (CS) of correspondences. These two steps are iterated until the largest CS is achieved [31]. Besides this, the least median squares (LMedS) [32] and the fast least trimmed squares (Fast-LTS) [33] have also been used [4, 34, 35]. There are also some other approaches which use different matching and retrieval algorithms with different features, which can be referred to the related reviewing

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

DOI: http://dx.doi.org/10.5772/intechopen.81665

Although lots of approaches have been developed for feature-based SAR image

registration, there are still some open problems that have not been perfectly solved yet. In this chapter, we concentrate on two problems, i.e., which feature is more appropriate and which retrieval algorithm performs much better? The first problem is related to the feature operator, which is focused in Sections 2 and 3. We give a detailed evaluation to tie points, Harris corner, SIFT, and SURF in terms of the geometrical invariance of feature and descriptor, extraction and matching speed, localization accuracy, robustness to decorrelation, and flexibility to speckle. SURF is identified to outperform others. Particularly, we find that SURF is flexible to speckle for the close relationship between Fast-Hessian detector and refined Lee speckle filter. SURF is thus more competent for SAR image registration. The second problem is posed in Section 4 with the reason that the widely used RANSAC is found instable for parameter estimation in the registration of an interferometric SAR (InSAR) image pair. The uncertainty arises from its inappropriate loss function and estimation strategy. Based on the scheme of Fast-LTS, an extended Fast-LTS (EF-LTS) is presented for 2D robust parameter estimation. Experiment on InSAR image pair demonstrates that EF-LTS is more stable and robust than RANSAC. It is more appropriate and competent for SAR image registration. Based on these, we recommend fitting the SURF features with EF-LTS to

conduct the registration. We further evaluate this scheme in Section 5 by

2. Comparative analysis on the commonly used features

Section 6 concludes the chapter finally.

for SAR image registration

Section 3.

17

processing the MiniSAR image pair, and the result complies with our expectation.

SAR image is acquired with intensity and phase, which should be transformed into the real one before feature detection by taking the intensity or the logarithmic intensity of the image. Instead of proposing a novel feature for SAR image registration, we identify the appropriate feature from the widely used tie points, Harris corner, SIFT, and SURF by evaluating them on several criteria. In this section, the features will be evaluated on the following six factors, i.e., the geometrical invariance of feature, the extraction speed, the localization accuracy, the geometric invariance of descriptor, the matching speed, and the robustness to decorrelation, while the impact of SAR speckles will be particularly focused and analyzed in

articles [36–38].

as their combination are often used for registration of multi-modality images. For SAR images with geometrical distortion and speckle, point feature is generally much clearer and easier extracted. Tie points, corner, and keypoint are the commonly used features in SAR image registration. Tie points usually refer to the features extracted from tie patches in SAR image registration [1–4]. The tie patches are first matched by region-based algorithms, and the tie points are then located by extracting the geometrical centers or centroids of the matched patches. Corner denotes another kind of point feature which has two dominant but different edge directions in local neighborhood. In SAR image registration, Harris corner [5] is the commonly used point feature [2, 6] whose response function is the weighted addition of the determinant and squared trace of the first-order moment matrix which describes the local neighboring gradient distribution of a point. Keypoint refers to the point differing in brightness or color compared with the surrounding. It is identified to further enable a complementary description of image structure that cannot be characterized by corner. The scale invariant feature transform (SIFT) [7] and the speeded up robust feature (SURF) [8] are the widely used keypoints in SAR image registration. SIFT was developed by Lowe [7] to extract features based on the automatic scale selection theory. Lindeberg [9] found that the only possible scalespace kernel under a variety of reasonable assumptions is the Gaussian function, and he experimented with both the traces of Hessian matrix, i.e., the Laplacian of Gaussian (LoG) and the determinant of Hessian (DoH) matrix, to detect the bloblike structures. To extract keypoints efficiently, Lowe [7] simplified LoG with the difference of Gaussian (DoG) further. SIFT enables not only a feature detector, but also a 128D vectorized descriptor of gradient and orientation. Mikolajczyk and Schmid conducted a comparative study on 10 different local descriptors and found that SIFT performs the best on treating the common image deformations [10]. SIFT has been widely used in SAR image registration [11–23]. Chen et al. [13] systematically evaluated the application of SIFT to SAR and displayed its usefulness for image registration. Schwind et al. [15] further indicated that SIFT is a robust alternative for point feature-based SAR image registration. The bottleneck of SIFT is the speed [8, 13, 15], which hinders its application to general SAR image registration. To accelerate SIFT, Schwind et al. [15] proposed to skip features detected at the first octave of the scale space pyramid (SSP) because matches extracted from this octave have the highest matching false alarm rate (MFAR). This can save the processing time without reducing the number of correct matches greatly. However, the first scale octave in SSP of SIFT refers to the image of original size or doubled size which has the highest resolution in SSP. Thus, the features extracted from this octave are more accurate for image registration [16]. Therefore, the discarding of matches from the first octave may influence the final registration accuracy. Based on the same scheme as SIFT, SURF developed by Bay et al. [8] uses a combination of novel detection, description, and matching methods to simplify SIFT. SURF extracts feature based on DoH instead of its trace because DoH bears slightly better scale selection property under non-Euclidean affine transformation than LoG. Bay et al. used a Fast-Hessian detector with box filters to approximate DoH. The SURF descriptor is a 64D vector composed by the Harr wavelet responses of the square area around keypoint. SURF has been demonstrated to outperform SIFT on speed, repeatability, distinctiveness, and robustness [8]. It has been used for multispectral satellite image registration [24], seabed recognition based on sonar images [25], and SAR image registration [26–29].

The next procedure after feature extraction is to match the features for correspondences. For tie points, this procedure is unnecessary because they have already matched when extracted. For other features, the correspondences are usually constructed by optimizing certain merit function, such as maximizing the similarity

#### On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

or minimizing the difference. The warp function can then be retrieved by fitting the obtained correspondences. For correspondences without any mismatches, the retrieval can be easily conducted by fitting them with the least squares (LS). However, for the general registration cases, the initial correspondences often contain mismatches. Therefore, the robust retrieval algorithms which are insensitive to outliers are needed. In many existing literatures on feature-based SAR image registration [15, 16, 26, 27], the random sample consensus (RANSAC) [30] has been widely used and recommended for warp function retrieval. RANSAC conducts the estimation by randomly sampling a minimal sampling set (MSS) to achieve an estimation of the warping, and the entire datasets are then checked on the estimation for a consensus set (CS) of correspondences. These two steps are iterated until the largest CS is achieved [31]. Besides this, the least median squares (LMedS) [32] and the fast least trimmed squares (Fast-LTS) [33] have also been used [4, 34, 35]. There are also some other approaches which use different matching and retrieval algorithms with different features, which can be referred to the related reviewing articles [36–38].

Although lots of approaches have been developed for feature-based SAR image registration, there are still some open problems that have not been perfectly solved yet. In this chapter, we concentrate on two problems, i.e., which feature is more appropriate and which retrieval algorithm performs much better? The first problem is related to the feature operator, which is focused in Sections 2 and 3. We give a detailed evaluation to tie points, Harris corner, SIFT, and SURF in terms of the geometrical invariance of feature and descriptor, extraction and matching speed, localization accuracy, robustness to decorrelation, and flexibility to speckle. SURF is identified to outperform others. Particularly, we find that SURF is flexible to speckle for the close relationship between Fast-Hessian detector and refined Lee speckle filter. SURF is thus more competent for SAR image registration. The second problem is posed in Section 4 with the reason that the widely used RANSAC is found instable for parameter estimation in the registration of an interferometric SAR (InSAR) image pair. The uncertainty arises from its inappropriate loss function and estimation strategy. Based on the scheme of Fast-LTS, an extended Fast-LTS (EF-LTS) is presented for 2D robust parameter estimation. Experiment on InSAR image pair demonstrates that EF-LTS is more stable and robust than RANSAC. It is more appropriate and competent for SAR image registration. Based on these, we recommend fitting the SURF features with EF-LTS to conduct the registration. We further evaluate this scheme in Section 5 by processing the MiniSAR image pair, and the result complies with our expectation. Section 6 concludes the chapter finally.

## 2. Comparative analysis on the commonly used features for SAR image registration

SAR image is acquired with intensity and phase, which should be transformed into the real one before feature detection by taking the intensity or the logarithmic intensity of the image. Instead of proposing a novel feature for SAR image registration, we identify the appropriate feature from the widely used tie points, Harris corner, SIFT, and SURF by evaluating them on several criteria. In this section, the features will be evaluated on the following six factors, i.e., the geometrical invariance of feature, the extraction speed, the localization accuracy, the geometric invariance of descriptor, the matching speed, and the robustness to decorrelation, while the impact of SAR speckles will be particularly focused and analyzed in Section 3.

as their combination are often used for registration of multi-modality images. For SAR images with geometrical distortion and speckle, point feature is generally much clearer and easier extracted. Tie points, corner, and keypoint are the commonly used features in SAR image registration. Tie points usually refer to the features extracted from tie patches in SAR image registration [1–4]. The tie patches are first matched by region-based algorithms, and the tie points are then located by extracting the geometrical centers or centroids of the matched patches. Corner denotes another kind of point feature which has two dominant but different edge directions in local neighborhood. In SAR image registration, Harris corner [5] is the commonly used point feature [2, 6] whose response function is the weighted addition of the determinant and squared trace of the first-order moment matrix which describes the local neighboring gradient distribution of a point. Keypoint refers to the point differing in brightness or color compared with the surrounding. It is identified to further enable a complementary description of image structure that cannot be characterized by corner. The scale invariant feature transform (SIFT) [7] and the speeded up robust feature (SURF) [8] are the widely used keypoints in SAR image registration. SIFT was developed by Lowe [7] to extract features based on the automatic scale selection theory. Lindeberg [9] found that the only possible scalespace kernel under a variety of reasonable assumptions is the Gaussian function, and he experimented with both the traces of Hessian matrix, i.e., the Laplacian of Gaussian (LoG) and the determinant of Hessian (DoH) matrix, to detect the bloblike structures. To extract keypoints efficiently, Lowe [7] simplified LoG with the difference of Gaussian (DoG) further. SIFT enables not only a feature detector, but also a 128D vectorized descriptor of gradient and orientation. Mikolajczyk and Schmid conducted a comparative study on 10 different local descriptors and found that SIFT performs the best on treating the common image deformations [10]. SIFT has been widely used in SAR image registration [11–23]. Chen et al. [13] systematically evaluated the application of SIFT to SAR and displayed its usefulness for image registration. Schwind et al. [15] further indicated that SIFT is a robust alternative for point feature-based SAR image registration. The bottleneck of SIFT is the speed [8, 13, 15], which hinders its application to general SAR image registration. To accelerate SIFT, Schwind et al. [15] proposed to skip features detected at the first octave of the scale space pyramid (SSP) because matches extracted from this octave have the highest matching false alarm rate (MFAR). This can save the processing time without reducing the number of correct matches greatly. However, the first scale octave in SSP of SIFT refers to the image of original size or doubled size which has the highest resolution in SSP. Thus, the features extracted from this octave are more accurate for image registration [16]. Therefore, the discarding of matches from the first octave may influence the final registration accuracy. Based on the same scheme as SIFT, SURF developed by Bay et al. [8] uses a combination of novel detection, description, and matching methods to simplify SIFT. SURF extracts feature based on DoH instead of its trace because DoH bears slightly better scale selection property under non-Euclidean affine transformation than LoG. Bay et al. used a Fast-Hessian detector with box filters to approximate DoH. The SURF descriptor is a 64D vector composed by the Harr wavelet responses of the square area around keypoint. SURF has been demonstrated to outperform SIFT on speed, repeatability, distinctiveness, and robustness [8]. It has been used for multispectral satellite image registration [24], seabed recognition based on sonar images [25], and

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

SAR image registration [26–29].

16

The next procedure after feature extraction is to match the features for correspondences. For tie points, this procedure is unnecessary because they have already matched when extracted. For other features, the correspondences are usually constructed by optimizing certain merit function, such as maximizing the similarity

#### 2.1 Geometrical invariance of feature

The geometrical invariance of feature refers to which degree of warping a same feature can still be extracted from the warped images by a detector. Crosscorrelation (CC) is sensitive to image rotation and scaling, hence the CC-based tie points are only invariant to the following translation transformation:

$$
\begin{bmatrix} x' \\ y' \\ \mathbf{1} \end{bmatrix} = \begin{bmatrix} \mathbf{1} & \mathbf{0} & t\_x \\ \mathbf{0} & \mathbf{1} & t\_y \\ \mathbf{0} & \mathbf{0} & \mathbf{1} \end{bmatrix} \begin{bmatrix} \times \\ y \\ \mathbf{1} \end{bmatrix} \tag{1}
$$

scale-invariant detector. For general SAR image application, scale-invariant features

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

The extraction speed is mainly influenced by the computational load of detector.

DoHSIFT <sup>¼</sup> Lxxð Þ <sup>x</sup>; <sup>σ</sup> Lyyð Þ� <sup>x</sup>; <sup>σ</sup> Lxyð Þ <sup>x</sup>; <sup>σ</sup> <sup>2</sup> (7)

Tie points are identified by traversing all potential offsets to calculate CC. The resulted computational load is heavy. The Harris response R is determined by the determinant and trace of matrix H. The calculation of H only relates to the firstorder derivatives which can be fast achieved. The scale-invariant SIFT and SURF keypoints are extracted by constructing SSP first. SSP is comprised of several octaves and each octave consists of several scale levels further. A scale level is a Gaussian-smoothed image. The nearby two layers are subtracted to calculate DoG, an approximation to LoG. The keypoint is finally identified as the point with extreme value of DoG in a 3 � 3 � 3 neighborhood in the scale space. SIFT detector performs slower than Harris because it extracts the feature in 3D space not in 2D space. Nonetheless, to extract the same number of subpixel features, SIFT detector is faster than CC-based tie points for the latter conducts exhaustive searching. SURF extracts feature based on DoH. Given a point x = (x, y) in image I at scale σ, the

where Lxx (x, σ), Lyy (x, σ), and Lxy (x, σ) denote the convolution of the Gaussian second-order derivative in x-, y-, and xy-directions with I, respectively. When applied in practice, Gaussians should be discretized and cropped. The corresponding discretized and cropped Lxx, Lxy, and Lyy with the lowest scale of 1.2 are displayed in the first row of Figure 1. Encouraged by the successful simplification of LoG with DoG in SIFT, Bay et al. devised a Fast-Hessian detector to approximate Lxx, Lxy, and Lyy with box filters Dxx, Dxy, and Dyy, respectively, shown in the second row of Figure 1. In [8], Bay et al. indicated that the performance of this approximation is comparable or even better than the original Gaussians. The

SIFT discretized and cropped Gaussian second-order partial derivatives in x- (Lxx), xy- (Lxy), and y-direction (Lyy), as well as their corresponding SURF box filter approximations Dxx , Dxy , and Dyy , respectively.

such as SIFT and SURF are sufficient.

DOI: http://dx.doi.org/10.5772/intechopen.81665

2.2 Feature extraction speed

scale function DoH is obtained by:

Figure 1.

19

where (x, <sup>y</sup>, 1)T \$ (x<sup>0</sup> , y<sup>0</sup> , 1)T are the inhomogeneous coordinates of a pair of matching points (the superscript T shows the vector transpose), and tx and ty denote the translations in x- and y-direction, respectively. The Harris measure is the following Harris matrix H describing the neighboring gradient distribution of a point [5]:

$$\mathbf{H} = \begin{bmatrix} \left< I\_x^2 \right> & \left< I\_x I\_y \right> \\ \left< I\_x I\_y \right> & \left< I\_y^2 \right> \end{bmatrix} \tag{2}$$

where <�> denotes the ensemble average; Ix and Iy are the first-order partial derivatives in x- and y-direction, respectively. Then, the response function R of Harris is the weighted sum of the determinant and squared trace of H [5]:

$$R = \det(\mathbf{H}) - \kappa (\text{trace}(\mathbf{H}))^2 \tag{3}$$

where the weight κ is a constant within the interval 0.04–0.06. A pixel is selected as a Harris corner if its response R is beyond a given threshold. It can be easily obtained from (2) that H is semi-definite Hermitian, which indicates the existence of two nonnegative eigenvalues λ<sup>1</sup> and λ2. Then (3) can be further formulated as:

$$R = \lambda\_1 \lambda\_2 - \kappa (\lambda\_1 + \lambda\_2)^2 \tag{4}$$

The Harris response R is only decided by the eigenvalues of H. Any unitary transformation of H will not influence the extraction of corner. Therefore, Harris corner is invariant to the following Euclidean transformation:

$$
\begin{bmatrix} x' \\ y' \\ \mathbf{1} \end{bmatrix} = \begin{bmatrix} \cos \theta & -\sin \theta & t\_{\mathbf{x}} \\ \sin \theta & \cos \theta & t\_{\mathbf{y}} \\ \mathbf{0} & \mathbf{0} & \mathbf{1} \end{bmatrix} \begin{bmatrix} \mathbf{x} \\ \mathbf{y} \\ \mathbf{1} \end{bmatrix} \tag{5}
$$

where θ denotes the rotation. SIFT and SURF were proposed to achieve the scale-invariance further:

$$
\begin{bmatrix} x' \\ y' \\ \mathbf{1} \end{bmatrix} = \begin{bmatrix} s\cos\theta & -s\sin\theta & t\_x \\ s\sin\theta & s\cos\theta & t\_y \\ \mathbf{0} & \mathbf{0} & \mathbf{1} \end{bmatrix} \begin{bmatrix} \mathbf{x} \\ y \\ \mathbf{1} \end{bmatrix} \tag{6}
$$

where s is the scale. Theoretically, SIFT and SURF features are not affineinvariant as Harris-Affine and Hessian-Affine features [39]. Nonetheless, the affine frame in Hessian-Affine and Harris-Affine is more sensitive to noise than scale-invariant detector. For general SAR image application, scale-invariant features such as SIFT and SURF are sufficient.

## 2.2 Feature extraction speed

2.1 Geometrical invariance of feature

where (x, <sup>y</sup>, 1)T \$ (x<sup>0</sup>

The geometrical invariance of feature refers to which degree of warping a same

matching points (the superscript T shows the vector transpose), and tx and ty denote the translations in x- and y-direction, respectively. The Harris measure is the following Harris matrix H describing the neighboring gradient distribution of a point [5]:

> I 2 x � � IxIy � �

2 4

IxIy � � I

where <�> denotes the ensemble average; Ix and Iy are the first-order partial derivatives in x- and y-direction, respectively. Then, the response function R of Harris is the weighted sum of the determinant and squared trace of H [5]:

where the weight κ is a constant within the interval 0.04–0.06. A pixel is selected

as a Harris corner if its response R is beyond a given threshold. It can be easily obtained from (2) that H is semi-definite Hermitian, which indicates the existence of two nonnegative eigenvalues λ<sup>1</sup> and λ2. Then (3) can be further formulated as:

R ¼ λ1λ<sup>2</sup> � κ λð Þ <sup>1</sup> þ λ<sup>2</sup>

The Harris response R is only decided by the eigenvalues of H. Any unitary transformation of H will not influence the extraction of corner. Therefore, Harris

where θ denotes the rotation. SIFT and SURF were proposed to achieve the

where s is the scale. Theoretically, SIFT and SURF features are not affineinvariant as Harris-Affine and Hessian-Affine features [39]. Nonetheless, the affine frame in Hessian-Affine and Harris-Affine is more sensitive to noise than

s cos θ �ssin θ tx ssin θ s cos θ ty 0 01

cos θ � sin θ tx sin θ cos θ ty 0 01

corner is invariant to the following Euclidean transformation:

2 6 4

x0 y0 1

x0 y0 1

2 6 4

2 6 4

2 6 4

scale-invariance further:

18

3 7 5

2 y D E

2 6 4

x y 1

3 7

, 1)T are the inhomogeneous coordinates of a pair of

3

<sup>R</sup> <sup>¼</sup> detð Þ� <sup>H</sup> <sup>κ</sup>ð Þ traceð Þ <sup>H</sup> <sup>2</sup> (3)

3 7 5

> 3 7 5

2 6 4

x y 1 3 7

2 6 4

x y 1 3 7

<sup>5</sup> (1)

5 (2)

<sup>2</sup> (4)

<sup>5</sup> (5)

<sup>5</sup> (6)

feature can still be extracted from the warped images by a detector. Crosscorrelation (CC) is sensitive to image rotation and scaling, hence the CC-based tie

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

2 6 4

points are only invariant to the following translation transformation:

x0 y0 1

H ¼

2 6 4

, y<sup>0</sup>

The extraction speed is mainly influenced by the computational load of detector. Tie points are identified by traversing all potential offsets to calculate CC. The resulted computational load is heavy. The Harris response R is determined by the determinant and trace of matrix H. The calculation of H only relates to the firstorder derivatives which can be fast achieved. The scale-invariant SIFT and SURF keypoints are extracted by constructing SSP first. SSP is comprised of several octaves and each octave consists of several scale levels further. A scale level is a Gaussian-smoothed image. The nearby two layers are subtracted to calculate DoG, an approximation to LoG. The keypoint is finally identified as the point with extreme value of DoG in a 3 � 3 � 3 neighborhood in the scale space. SIFT detector performs slower than Harris because it extracts the feature in 3D space not in 2D space. Nonetheless, to extract the same number of subpixel features, SIFT detector is faster than CC-based tie points for the latter conducts exhaustive searching. SURF extracts feature based on DoH. Given a point x = (x, y) in image I at scale σ, the scale function DoH is obtained by:

$$
\Delta DoH\_{\rm SIFT} = L\_{\rm xx}(\mathbf{x}, \sigma) L\_{\rm \mathcal{Y}}(\mathbf{x}, \sigma) - \left( L\_{\rm xy}(\mathbf{x}, \sigma) \right)^2 \tag{7}
$$

where Lxx (x, σ), Lyy (x, σ), and Lxy (x, σ) denote the convolution of the Gaussian second-order derivative in x-, y-, and xy-directions with I, respectively.

When applied in practice, Gaussians should be discretized and cropped. The corresponding discretized and cropped Lxx, Lxy, and Lyy with the lowest scale of 1.2 are displayed in the first row of Figure 1. Encouraged by the successful simplification of LoG with DoG in SIFT, Bay et al. devised a Fast-Hessian detector to approximate Lxx, Lxy, and Lyy with box filters Dxx, Dxy, and Dyy, respectively, shown in the second row of Figure 1. In [8], Bay et al. indicated that the performance of this approximation is comparable or even better than the original Gaussians. The

#### Figure 1.

SIFT discretized and cropped Gaussian second-order partial derivatives in x- (Lxx), xy- (Lxy), and y-direction (Lyy), as well as their corresponding SURF box filter approximations Dxx , Dxy , and Dyy , respectively.

approximation makes pixels in certain window have the same weight. The convolutions can be then calculated at very low computational cost by using the integral image. Therefore, instead of iteratively reducing the image size and using the cascade filtering, SSP in SURF is built by simply up-scaling the box filters without changing the size of the image. The use of integral image enables the convolutions independent of the filter size and scale.

maximize the similarity (such as CC [4]) or minimize the differences (such as Euclidean distance [7, 8]). A correspondence is detected if it can optimize the merit function. For SIFT and SURF, the merit of an optimal correspondence has also to be certain times larger than the second optimal merit. Matching speed is mainly determined by the calculation of merit. For tie points and Harris corner, the merit function is the maximum of CC, which can be obtained on complex data or magnitude data [44], referring to coherent CC or incoherent CC, respectively. The registration accuracy attained by coherent CC is much higher than that by incoherent CC [45]. If D<sup>1</sup> and D<sup>2</sup> are the image patches, respectively, centered at an initial

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

<sup>j</sup>¼<sup>1</sup> <sup>D</sup>1ð Þ� <sup>i</sup>; <sup>j</sup> <sup>μ</sup><sup>1</sup> ð Þ <sup>D</sup>2ð Þ� <sup>i</sup>; <sup>j</sup> <sup>μ</sup><sup>2</sup> ð Þ<sup>∗</sup> �

<sup>j</sup>¼<sup>1</sup> <sup>D</sup>1ð Þ� <sup>i</sup>; <sup>j</sup> <sup>μ</sup><sup>1</sup> j j ð Þ <sup>2</sup>

Dist Dð Þ¼ <sup>3</sup>; D<sup>4</sup> ∑

where N is the size of the image patch, μ<sup>1</sup> and μ<sup>1</sup> denote the means of D<sup>1</sup> and D2, respectively. Equation (10) requires about 10N<sup>2</sup> operations including 7N<sup>2</sup>

> L i¼1

where L is the length of descriptor. Equation (11) requires 3L operations including 2L additions and L multiplications. For SURF, Bay et al. [8] found that the sign of Laplacian can be further used to distinguish the feature from its background for fast indexing during matching stage. The merit will not be computed unless the initial match has the same sign. Hence, under the assumption of equal probability distribution for sign of Laplacian, the merit computation in SURF requires 1.5L operations. Taking the descriptor lengths L for SIFT and SURF being 128 and 64 into consideration, then (11) involves in 384 and 96 operations for SIFT and SURF, respectively. Hence, SURF is four times faster than SIFT on feature matching. To achieve the same efficiency as SIFT or SURF, the equivalent patch size N for tie points and Harris corner should be about 6 or 3, respectively. This may lead to biased CC estimation thus bad feature localization and matching due to the insufficient

SAR decorrelation sources can be classified into two categories, i.e., the geometrical warping and radiometric warping. Geometrical warping will lead to decorrelation and influence the CC-based feature matching, which relates to the geometrical invariance of feature discussed above. Here, we focus on the

CC is only invariant to affine changes in scattering. Target scattering in

radiometric warping-induced decorrelation. Such decorrelation is resulted because

microwave band is sensitive to frequency, bandwidth, and polarization. All these introduce a complex nonlinear radiometric warping, which degrades SAR

information and aggravates image registration by impacting the localization of tie points. The localization accuracy of tie points is measured by the error standard

The merit function in SIFT and SURF is the minimum of the Euclidean distance. If D<sup>3</sup> and D<sup>4</sup> are the descriptors of an initial match, respectively, the distance can

� ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

∑<sup>N</sup> <sup>i</sup>¼<sup>1</sup>∑<sup>N</sup> <sup>j</sup>¼<sup>1</sup> <sup>D</sup>2ð Þ� <sup>i</sup>; <sup>j</sup> <sup>μ</sup><sup>2</sup> j j ð Þ <sup>2</sup> <sup>q</sup> (10)

� �

j j <sup>D</sup>3ðÞ�<sup>i</sup> <sup>D</sup>4ð Þ<sup>i</sup> <sup>2</sup> (11)

match, the coherent CC is calculated as

DOI: http://dx.doi.org/10.5772/intechopen.81665

additions and 3N<sup>2</sup> multiplications.

2.6 Robustness to decorrelation

∑<sup>N</sup> <sup>i</sup>¼<sup>1</sup>∑<sup>N</sup>

� �

∑<sup>N</sup> <sup>i</sup>¼<sup>1</sup>∑<sup>N</sup>

CC Dð Þ¼ <sup>1</sup>; D<sup>2</sup>

be calculated by

sampling.

deviation σ<sup>L</sup> [45]:

21

#### 2.3 Localization accuracy of feature

Image registration accuracy is closely determined by the localization accuracy of feature. Tie points achieve subpixel accuracy by oversampling the image patches [40] or CC obtained in coarse registration [41]. Higher sampling rate indicates higher accuracy, but it also signifies larger data sets, heavier computational load, and more severe aliasing. Keypoint in SIFT and SURF is first located as the extrema using the non-maximum suppression technique, and is then refined to subpixel and sub-scale accuracy by Taylor fitting a 3D quadratic to the scale function DoG (for SIFT) or the approximated DoH (for SURF) in the scale space [42]:

$$f(\mathbf{X}) = f(\mathbf{X}\_0) + \left(\frac{\partial f}{\partial \mathbf{X}}(\mathbf{X}\_0)\right)^\mathrm{T} \mathbf{A} \mathbf{X} + \frac{1}{2} \Delta \mathbf{X}^\mathrm{T} \left(\frac{\partial^2 f}{\partial \mathbf{x}^2}(\mathbf{X}\_0)\right) \mathbf{A} \mathbf{X}.\tag{8}$$

Therefore, SIFT and SURF can obtain the highest accuracy. However, it should be noted that although the subpixel feature localization is the precondition of accurate image registration, it cannot guarantee a subpixel image registration. For high accurate SAR image registration, we should further evaluate the features carefully, and this will be detailed in Section 3.4.

#### 2.4 Geometrical invariance of descriptor

Feature descriptor is usually a vector depicting the neighboring information of a feature. It plays a key role in feature matching. The descriptor's geometrical invariance determines the degree of warping to which features can still be successfully matched. Harris corner and tie points have no descriptor. From feature matching point of view, however, they both adopt template matching by selecting the image square centered around the feature as descriptor, which is only invariant to translation. Thus, tie points and Harris corner can be successfully matched only under weak warping. SIFT and SURF descriptors enable a good compromise between feature complexity and the robustness to commonly occurring deformation such as weak affine transformation [7, 8, 43]:

$$
\begin{bmatrix} x' \\ y' \\ \mathbf{1} \end{bmatrix} = \begin{bmatrix} s\_{\mathbf{x}} \cos \theta & -s\_{\mathbf{y}} \sin \theta & t\_{\mathbf{x}} \\ s\_{\mathbf{x}} \sin \theta & s\_{\mathbf{y}} \cos \theta & t\_{\mathbf{y}} \\ \mathbf{0} & \mathbf{0} & \mathbf{1} \end{bmatrix} \begin{bmatrix} \mathbf{x} \\ y \\ \mathbf{1} \end{bmatrix} \tag{9}
$$

where sx and sy denote the scales in directions x and y, respectively. Robust matching across a substantial range of affine distortion and change in 3D viewpoint can hence be achieved.

#### 2.5 Matching speed of feature

Feature matching is usually conducted based on certain merit function of the descriptors. In feature-based SAR image registration, the merit function is to

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

maximize the similarity (such as CC [4]) or minimize the differences (such as Euclidean distance [7, 8]). A correspondence is detected if it can optimize the merit function. For SIFT and SURF, the merit of an optimal correspondence has also to be certain times larger than the second optimal merit. Matching speed is mainly determined by the calculation of merit. For tie points and Harris corner, the merit function is the maximum of CC, which can be obtained on complex data or magnitude data [44], referring to coherent CC or incoherent CC, respectively. The registration accuracy attained by coherent CC is much higher than that by incoherent CC [45]. If D<sup>1</sup> and D<sup>2</sup> are the image patches, respectively, centered at an initial match, the coherent CC is calculated as

$$\text{CC}(D\_1, D\_2) = \frac{\left| \sum\_{i=1}^{N} \sum\_{j=1}^{N} (D\_1(i, j) - \mu\_1)(D\_2(i, j) - \mu\_2)^\* \right|}{\sqrt{\sum\_{i=1}^{N} \sum\_{j=1}^{N} |(D\_1(i, j) - \mu\_1)|^2 \sum\_{i=1}^{N} \sum\_{j=1}^{N} |(D\_2(i, j) - \mu\_2)|^2}} \tag{10}$$

where N is the size of the image patch, μ<sup>1</sup> and μ<sup>1</sup> denote the means of D<sup>1</sup> and D2, respectively. Equation (10) requires about 10N<sup>2</sup> operations including 7N<sup>2</sup> additions and 3N<sup>2</sup> multiplications.

The merit function in SIFT and SURF is the minimum of the Euclidean distance. If D<sup>3</sup> and D<sup>4</sup> are the descriptors of an initial match, respectively, the distance can be calculated by

$$Dist(D\_3, D\_4) = \sum\_{i=1}^{L} |D\_3(i) - D\_4(i)|^2 \tag{11}$$

where L is the length of descriptor. Equation (11) requires 3L operations including 2L additions and L multiplications. For SURF, Bay et al. [8] found that the sign of Laplacian can be further used to distinguish the feature from its background for fast indexing during matching stage. The merit will not be computed unless the initial match has the same sign. Hence, under the assumption of equal probability distribution for sign of Laplacian, the merit computation in SURF requires 1.5L operations. Taking the descriptor lengths L for SIFT and SURF being 128 and 64 into consideration, then (11) involves in 384 and 96 operations for SIFT and SURF, respectively. Hence, SURF is four times faster than SIFT on feature matching. To achieve the same efficiency as SIFT or SURF, the equivalent patch size N for tie points and Harris corner should be about 6 or 3, respectively. This may lead to biased CC estimation thus bad feature localization and matching due to the insufficient sampling.

#### 2.6 Robustness to decorrelation

SAR decorrelation sources can be classified into two categories, i.e., the geometrical warping and radiometric warping. Geometrical warping will lead to decorrelation and influence the CC-based feature matching, which relates to the geometrical invariance of feature discussed above. Here, we focus on the radiometric warping-induced decorrelation. Such decorrelation is resulted because CC is only invariant to affine changes in scattering. Target scattering in microwave band is sensitive to frequency, bandwidth, and polarization. All these introduce a complex nonlinear radiometric warping, which degrades SAR information and aggravates image registration by impacting the localization of tie points. The localization accuracy of tie points is measured by the error standard deviation σ<sup>L</sup> [45]:

approximation makes pixels in certain window have the same weight. The convolutions can be then calculated at very low computational cost by using the integral image. Therefore, instead of iteratively reducing the image size and using the cascade filtering, SSP in SURF is built by simply up-scaling the box filters without changing the size of the image. The use of integral image enables the convolutions

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Image registration accuracy is closely determined by the localization accuracy of feature. Tie points achieve subpixel accuracy by oversampling the image patches [40] or CC obtained in coarse registration [41]. Higher sampling rate indicates higher accuracy, but it also signifies larger data sets, heavier computational load, and more severe aliasing. Keypoint in SIFT and SURF is first located as the extrema using the non-maximum suppression technique, and is then refined to subpixel and sub-scale accuracy by Taylor fitting a 3D quadratic to the scale function DoG (for

> ΔX þ 1 2

Therefore, SIFT and SURF can obtain the highest accuracy. However, it should

Feature descriptor is usually a vector depicting the neighboring information of a feature. It plays a key role in feature matching. The descriptor's geometrical invariance determines the degree of warping to which features can still be successfully matched. Harris corner and tie points have no descriptor. From feature matching point of view, however, they both adopt template matching by selecting the image square centered around the feature as descriptor, which is only invariant to translation. Thus, tie points and Harris corner can be successfully matched only under weak warping. SIFT and SURF descriptors enable a good compromise between feature complexity and the robustness to commonly occurring deformation such as

> sx cos θ �sy sin θ tx sx sin θ sy cos θ ty 0 01

where sx and sy denote the scales in directions x and y, respectively. Robust matching across a substantial range of affine distortion and change in 3D viewpoint

Feature matching is usually conducted based on certain merit function of the descriptors. In feature-based SAR image registration, the merit function is to

3 7 5

2 6 4

x y 1 3 7

<sup>5</sup> (9)

be noted that although the subpixel feature localization is the precondition of accurate image registration, it cannot guarantee a subpixel image registration. For high accurate SAR image registration, we should further evaluate the features

<sup>Δ</sup>X<sup>T</sup> <sup>∂</sup><sup>2</sup>

f <sup>∂</sup>x<sup>2</sup> ð Þ <sup>X</sup><sup>0</sup> � �

ΔX: (8)

SIFT) or the approximated DoH (for SURF) in the scale space [42]:

∂f <sup>∂</sup><sup>X</sup> ð Þ <sup>X</sup><sup>0</sup> � �<sup>T</sup>

independent of the filter size and scale.

2.3 Localization accuracy of feature

fð Þ¼ X fð Þþ X<sup>0</sup>

carefully, and this will be detailed in Section 3.4.

2.4 Geometrical invariance of descriptor

weak affine transformation [7, 8, 43]:

can hence be achieved.

20

2.5 Matching speed of feature

x0 y0 1

2 6 4

2 6 4 Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

$$
\sigma\_L = \sqrt{\frac{3}{2\text{N}^2}} \frac{\sqrt{1-\chi^2}}{\pi\chi} \text{o} sr^{3/2} \tag{12}
$$

3.2 Refined Lee speckle filter

DOI: http://dx.doi.org/10.5772/intechopen.81665

without degrading the edge [46].

Figure 2.

23

The pixels in white are used for filtering computation.

in <sup>y</sup>-direction with convolution template [1 2 1]<sup>T</sup>

An ideal speckle filter should adaptively smooth speckle, retain the sharpness of boundaries and edges, and preserve the subtle but distinguishable details. The most widely used boxcar filter replaces a pixel with the mean of its windowed neighborhood. This filter can be easily implemented and works very well in homogeneous area, but will degrade spatial resolution in inhomogeneous area due to the indiscriminate averaging [46]. To solve this, many filtering techniques have been proposed. The refined Lee speckle filter is just such a filter which uses the local statistics to suppress speckles without degrading image. To identify pixels with the similar texture, Lee devised the eight non-square edge-aligned windows, as shown in Figure 2. In the course of filtering, one of the windows is matched to calculate local statistics based on edge direction, and the minimum mean square algorithm is then adopted for filtering. As a result, this filter can effectively reduce the speckle

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

3.3 Relationship between Fast-Hessian detector and refined Lee filter

As mentioned previously, SURF extracts features based on the box filter displayed in Figure 1. Box filter not only speeds up feature extraction, but also enables SURF to extract features while reducing speckles. In Dxx of Figure 1, we average the pixels using a 5 3 window first, and then extract the vertical edge by the second-order image partial derivative in x-direction with convolution template [1 2 1]. This is equivalent to filter speckles with Lee's windows Figure 2(a) and (e). Similarly, Dyy denotes that we also filter the pixels using a 5 3 window first, but then extract the horizontal edge using the second-order image partial derivative

speckle with Lee's non-square windows Figure 2(c) and (g). Dxy shows that we use a 3 3 window and extract the 135° edge feature by the second-order image partial derivative in negative xy-direction with the convolution template [1 1; 1, 1]. This is equivalent to filter speckle with windows Figure 2(d) and (h). Likewise, Dxy gives that we also use a 3 3 window but extract the 45° edge by the second order image partial derivative in positive xy-direction with convolution template [1 1; 1, 1]. This is equivalent to filter with windows Figure 2(b) and (f). Instead of

Edge-aligned windows used in refined Lee filter to decide the local texture, where windows (a) and (e) are used for vertical edge, (c) and (g) for horizontal edge, (b) and (f) for 135° edge, and (d) and (h) for 45° edge.

. This is equivalent to filter

where γ is CC, N is the size of the tie patches, and osr is the oversampling rate. Localization accuracy directly relates to CC: higher coherence means higher localization accuracy, while higher decorrelation indicates worse localization accuracy and worse registration accuracy. It is known that one can approximate a nonlinear function with a series of linear functions, so a nice method to improve the robustness to decorrelation is to use smaller image patches, but this will also result in worse localization accuracy through N in (12). Thus, tie points are not robust to decorrelation. Similarly, the influence of decorrelation on CC-based matching of Harris corners is also unavoidable. However, Harris, SIFT, and SURF locate feature based on geometrical texture instead of correlation. This will reduce the influence of decorrelation. The matching of SIFT and SURF features is based on local descriptors which are invariant to affine changes in scattering. SIFT and SURF features are thus more robust to decorrelation.

#### 3. Impact of SAR speckles on accurate feature extraction

SAR image is acquired by actively measuring and coherently processing the electromagnetic scattering of target. The interference of scatterings from scatterers within each resolution cell produces a pixel-to-pixel variation in image intensity and results in the so-called speckle. In this section, we first conduct a qualitative evaluation on the flexibility of existing features to speckles. An experimental evaluation of the identified feature is then conducted and some necessary improvements are developed for high accurate SAR image registration.

#### 3.1 Flexibility to image speckling

For CC-based tie points, the assumption that the scattering is locally stationary and ergodic may not be tenable in the existence of speckles. As a result, the correlation estimation as well as the localization and matching of the feature will be biased. For the geometrical texture-based detectors such as Harris, SIFT, and SURF, speckles may lead to false texture and high MFAR. To achieve stable features from the speckle-contaminated SAR image, a conceivable method is to suppress speckle beforehand. Schwind et al. [15] suggested adopting the ISEF filter, but they indicated that ISEF filter and any other filter may slightly affect feature localization and registration quality. Hence, a better strategy is to conduct speckle suppression while feature extraction, i.e., the detector should be flexible to speckling.

Harris detector obtains features using the first-order image derivatives which are not robust to speckles. As a result, Harris detector may extract many features, but most of the extracted features are speckles with only a few correct matches. This influence has been also observed by Schwind et al. [15] when using SIFT to SAR: only very few matches are constructed at the first octave of SSP although with extensive number of extractable features, and the matches from this octave have the highest MFAR of all the octaves. The first scale octave refers to the original or double-sized images which are of the highest resolution and the largest number of extractable keypoints. The highest MFAR at this octave clearly indicates the bad flexibility of SIFT to speckles, while the lower MFAR at higher octaves is just due to the fact that larger image smoothing reduces the speckle. Different from SIFT, SURF can deal with speckle very well because of the relationship between Fast-Hessian detector and refined Lee speckle filter.

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

## 3.2 Refined Lee speckle filter

σ<sup>L</sup> ¼

3. Impact of SAR speckles on accurate feature extraction

feature extraction, i.e., the detector should be flexible to speckling.

Hessian detector and refined Lee speckle filter.

22

developed for high accurate SAR image registration.

3.1 Flexibility to image speckling

SAR image is acquired by actively measuring and coherently processing the electromagnetic scattering of target. The interference of scatterings from scatterers within each resolution cell produces a pixel-to-pixel variation in image intensity and results in the so-called speckle. In this section, we first conduct a qualitative evaluation on the flexibility of existing features to speckles. An experimental evaluation of the identified feature is then conducted and some necessary improvements are

For CC-based tie points, the assumption that the scattering is locally stationary and ergodic may not be tenable in the existence of speckles. As a result, the correlation estimation as well as the localization and matching of the feature will be biased. For the geometrical texture-based detectors such as Harris, SIFT, and SURF, speckles may lead to false texture and high MFAR. To achieve stable features from the speckle-contaminated SAR image, a conceivable method is to suppress speckle beforehand. Schwind et al. [15] suggested adopting the ISEF filter, but they indicated that ISEF filter and any other filter may slightly affect feature localization and registration quality. Hence, a better strategy is to conduct speckle suppression while

Harris detector obtains features using the first-order image derivatives which are not robust to speckles. As a result, Harris detector may extract many features, but most of the extracted features are speckles with only a few correct matches. This influence has been also observed by Schwind et al. [15] when using SIFT to SAR: only very few matches are constructed at the first octave of SSP although with extensive number of extractable features, and the matches from this octave have the highest MFAR of all the octaves. The first scale octave refers to the original or double-sized images which are of the highest resolution and the largest number of extractable keypoints. The highest MFAR at this octave clearly indicates the bad flexibility of SIFT to speckles, while the lower MFAR at higher octaves is just due to the fact that larger image smoothing reduces the speckle. Different from SIFT, SURF can deal with speckle very well because of the relationship between Fast-

more robust to decorrelation.

ffiffiffiffiffiffiffiffi 3 2N<sup>2</sup>

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

r ffiffiffiffiffiffiffiffiffiffiffiffi <sup>1</sup> � <sup>γ</sup><sup>2</sup> <sup>p</sup> πγ

where γ is CC, N is the size of the tie patches, and osr is the oversampling rate. Localization accuracy directly relates to CC: higher coherence means higher localization accuracy, while higher decorrelation indicates worse localization accuracy and worse registration accuracy. It is known that one can approximate a nonlinear function with a series of linear functions, so a nice method to improve the robustness to decorrelation is to use smaller image patches, but this will also result in worse localization accuracy through N in (12). Thus, tie points are not robust to decorrelation. Similarly, the influence of decorrelation on CC-based matching of Harris corners is also unavoidable. However, Harris, SIFT, and SURF locate feature based on geometrical texture instead of correlation. This will reduce the influence of decorrelation. The matching of SIFT and SURF features is based on local descriptors which are invariant to affine changes in scattering. SIFT and SURF features are thus

osr3=<sup>2</sup> (12)

An ideal speckle filter should adaptively smooth speckle, retain the sharpness of boundaries and edges, and preserve the subtle but distinguishable details. The most widely used boxcar filter replaces a pixel with the mean of its windowed neighborhood. This filter can be easily implemented and works very well in homogeneous area, but will degrade spatial resolution in inhomogeneous area due to the indiscriminate averaging [46]. To solve this, many filtering techniques have been proposed. The refined Lee speckle filter is just such a filter which uses the local statistics to suppress speckles without degrading image. To identify pixels with the similar texture, Lee devised the eight non-square edge-aligned windows, as shown in Figure 2. In the course of filtering, one of the windows is matched to calculate local statistics based on edge direction, and the minimum mean square algorithm is then adopted for filtering. As a result, this filter can effectively reduce the speckle without degrading the edge [46].

#### 3.3 Relationship between Fast-Hessian detector and refined Lee filter

As mentioned previously, SURF extracts features based on the box filter displayed in Figure 1. Box filter not only speeds up feature extraction, but also enables SURF to extract features while reducing speckles. In Dxx of Figure 1, we average the pixels using a 5 3 window first, and then extract the vertical edge by the second-order image partial derivative in x-direction with convolution template [1 2 1]. This is equivalent to filter speckles with Lee's windows Figure 2(a) and (e). Similarly, Dyy denotes that we also filter the pixels using a 5 3 window first, but then extract the horizontal edge using the second-order image partial derivative in <sup>y</sup>-direction with convolution template [1 2 1]<sup>T</sup> . This is equivalent to filter speckle with Lee's non-square windows Figure 2(c) and (g). Dxy shows that we use a 3 3 window and extract the 135° edge feature by the second-order image partial derivative in negative xy-direction with the convolution template [1 1; 1, 1]. This is equivalent to filter speckle with windows Figure 2(d) and (h). Likewise, Dxy gives that we also use a 3 3 window but extract the 45° edge by the second order image partial derivative in positive xy-direction with convolution template [1 1; 1, 1]. This is equivalent to filter with windows Figure 2(b) and (f). Instead of

#### Figure 2.

Edge-aligned windows used in refined Lee filter to decide the local texture, where windows (a) and (e) are used for vertical edge, (c) and (g) for horizontal edge, (b) and (f) for 135° edge, and (d) and (h) for 45° edge. The pixels in white are used for filtering computation.

selecting the optimal edge to calculate local statistics, the four edge features are combined to a new feature in SURF by:

$$DoH\_{SURF} = D\_{\infty} D\_{\mathcal{Y}} + \left(0.9 D\_{\text{xy}}\right) \left(-0.9 D\_{\text{xy}}\right) \tag{13}$$

xs ys 1

3 7 <sup>5</sup> <sup>¼</sup> <sup>A</sup>

ATE <sup>¼</sup> <sup>1</sup>

<sup>N</sup> <sup>∑</sup> N i¼1 xsi ysi 1

where Ar indicates the warp matrix retrieved on all the constructed correspondences (xsi, ysi) and (xmi, ymi) denote the ith correct correspondence located in slave image and master image, respectively, and N is the number of correct matches

3 7 <sup>5</sup> � Ar

xmi ymi 1

3 7 5

� � � � � � �

2 6 4

<sup>&</sup>lt; threshold True xsi; ysi � � \$ xmi; ymi � � is a correct match False xsi; ysi � � \$ xmi; ymi � � is a mismatch (

2 6 4

� � � � � � �

xm ym 1

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

2 6 4

where (x, y, 1)<sup>T</sup> are the homogenous image coordinates, subscripts s and m denote the slave and master images, respectively. A is an affine matrix composed by parameters a, b, c, and d, as well as two translations tx and ty. Bay et al. devised two versions of Fast-Hessian detectors for SURF. The one initializes SSP by using 9 � 9 box filter to the original image is denoted as FH-9(-1), while the one initializes SSP by using 15 � 15 box filter to double-sized image (also with doubled sampling step) is denoted as FH-15(-2). FH-15(-2) has been shown to be better than FH-9(-1) on repeatability [8]. We use the two detectors to extract point correspondences, respectively, based on which the robust EF-LTS (will be presented in Section 4) is then used to retrieve the warp matrix. To compare the two SURF detectors for SAR image registration, we consider four criteria, i.e., the average transfer error (ATE), MFAR, the number of correct matches, and the warp matrix estimation error (WMEE). ATE measures the appropriateness of the extracted features to the

abtx cdty 00 1 3 7 5

2 6 4

xm ym 1

3 7

<sup>5</sup> (14)

(15)

(16)

2 6 4

2 6 4

DOI: http://dx.doi.org/10.5772/intechopen.81665

achieved warp parameters:

which are selected by:

xmi ymi 1

3 7 5

� � � � � � �

SAR image of Enta Volcano taken by SIR-C/X-SAR (300 � 300).

2 6 4

xsi ysi 1

Figure 3.

25

3 7 <sup>5</sup> � <sup>A</sup>

2 6 4

� � � � � � �

which corresponds to DoH in (7), where the constant 0.9 is used to balance the expression for the Hessian's determinant. Then, SSP in SURF just indicates that we adopt a series of box filters of different size to filter speckles and extract features of different scales. Hence, SURF is very flexible to deal with speckle.

## 3.4 Evaluation of SURF for SAR image subpixel registration

As listed in Table 1, according to the comparative analysis in Sections 2 and 3.1 on several criteria, we can obtain that for the general registration of SAR images


From these, we can see that SURF is more appropriate and competent for general SAR image registration. Nevertheless, SAR applications, like DEM retrieval and deformation estimation usually impose a strict requirement for registration accuracy. To ensure an acceptable result, the registration accuracy should be subpixel. To evaluate the capability of SURF for subpixel image registration, we devise a comparative experiment on some contrived SAR image pairs. Figure 3 shows a SAR image of Enta Volcano acquired by SIR-C/X-SAR. We treat this image as the master and transform it to model an affine geometrical warp for the slave image:


#### Table 1.

Evaluation of the four commonly used features for SAR image registration in terms of several criteria.

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

$$
\begin{bmatrix} \mathbf{x}\_{\boldsymbol{t}} \\ \mathbf{y}\_{\boldsymbol{t}} \\ \mathbf{1} \end{bmatrix} = \mathbf{A} \begin{bmatrix} \mathbf{x}\_{m} \\ \mathbf{y}\_{m} \\ \mathbf{1} \end{bmatrix} = \begin{bmatrix} a & b & t\_{\mathbf{x}} \\ c & d & t\_{\mathbf{y}} \\ \mathbf{0} & \mathbf{0} & \mathbf{1} \end{bmatrix} \begin{bmatrix} \mathbf{x}\_{m} \\ \mathbf{y}\_{m} \\ \mathbf{1} \end{bmatrix} \tag{14}$$

where (x, y, 1)<sup>T</sup> are the homogenous image coordinates, subscripts s and m denote the slave and master images, respectively. A is an affine matrix composed by parameters a, b, c, and d, as well as two translations tx and ty. Bay et al. devised two versions of Fast-Hessian detectors for SURF. The one initializes SSP by using 9 � 9 box filter to the original image is denoted as FH-9(-1), while the one initializes SSP by using 15 � 15 box filter to double-sized image (also with doubled sampling step) is denoted as FH-15(-2). FH-15(-2) has been shown to be better than FH-9(-1) on repeatability [8]. We use the two detectors to extract point correspondences, respectively, based on which the robust EF-LTS (will be presented in Section 4) is then used to retrieve the warp matrix. To compare the two SURF detectors for SAR image registration, we consider four criteria, i.e., the average transfer error (ATE), MFAR, the number of correct matches, and the warp matrix estimation error (WMEE). ATE measures the appropriateness of the extracted features to the achieved warp parameters:

$$ATE = \frac{1}{N} \sum\_{i=1}^{N} \left| \begin{bmatrix} \boldsymbol{\chi}\_{si} \\ \boldsymbol{\upnu}\_{si} \\ \mathbf{1} \end{bmatrix} - \mathbf{A}\_{\mathbf{r}} \begin{bmatrix} \boldsymbol{\upkappa}\_{mi} \\ \boldsymbol{\upnu}\_{mi} \\ \mathbf{1} \end{bmatrix} \right| \tag{15}$$

where Ar indicates the warp matrix retrieved on all the constructed correspondences (xsi, ysi) and (xmi, ymi) denote the ith correct correspondence located in slave image and master image, respectively, and N is the number of correct matches which are selected by:

$$\begin{array}{c|c} \begin{bmatrix} \mathbf{x}\_{si} \\ \mathbf{y}\_{si} \\ \mathbf{1} \end{bmatrix} - \mathbf{A} \begin{bmatrix} \mathbf{x}\_{mi} \\ \mathbf{y}\_{mi} \\ \mathbf{1} \end{bmatrix} \end{array} \text{threshold} \begin{cases} \begin{array}{c} \text{True} \\ \begin{cases} \mathbf{True} \\ \mathbf{False} \end{cases} \begin{array}{c} (\mathbf{x}\_{si}, \mathbf{y}\_{si}) \leftrightarrow (\mathbf{x}\_{mi}, \mathbf{y}\_{mi}) \text{ is a correct match} \\\end{array} \\\hline \begin{array}{c} (\mathbf{x}\_{si}, \mathbf{y}\_{si}) \leftrightarrow (\mathbf{x}\_{mi}, \mathbf{y}\_{mi}) \text{ is a mismatch} \end{array} \end{array} \end{array} \tag{4.6}$$

Figure 3. SAR image of Enta Volcano taken by SIR-C/X-SAR (300 � 300).

selecting the optimal edge to calculate local statistics, the four edge features are

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

which corresponds to DoH in (7), where the constant 0.9 is used to balance the expression for the Hessian's determinant. Then, SSP in SURF just indicates that we adopt a series of box filters of different size to filter speckles and extract features of

As listed in Table 1, according to the comparative analysis in Sections 2 and 3.1 on several criteria, we can obtain that for the general registration of SAR images

• Tie points are fit for images with slight distortion and weak decorrelation

From these, we can see that SURF is more appropriate and competent for general SAR image registration. Nevertheless, SAR applications, like DEM retrieval and deformation estimation usually impose a strict requirement for registration accuracy. To ensure an acceptable result, the registration accuracy should be subpixel. To evaluate the capability of SURF for subpixel image registration, we devise a comparative experiment on some contrived SAR image pairs. Figure 3 shows a SAR image of Enta Volcano acquired by SIR-C/X-SAR. We treat this image as the master and transform it to model an affine geometrical warp for the slave

Items Tie points Harris corner SIFT SURF

translation

Scaling, rotation, and translation

Subpixel\* Pixel Subpixel Subpixel

Worse Bad Good Good

Good Bad Bad Better

Translation Translation Affine transform Affine transform

Scaling, rotation, and translation

Translation Rotation and

Feature extraction speed Slower Faster Slow Fast

Feature matching speed Slow Slow Fast Faster

Evaluation of the four commonly used features for SAR image registration in terms of several criteria.

�0:9Dxy

(13)

DoHSURF ¼ DxxDyy þ 0:9Dxy

different scales. Hence, SURF is very flexible to deal with speckle.

3.4 Evaluation of SURF for SAR image subpixel registration

• SURF outperforms others in terms of the considered criteria.

• SIFT is applicable when no strict requirement for speed.

• Harris may be appropriate for coarse registration.

and require heavy computation load.

image:

feature

accuracy

Robustness to decorrelation

speckle

Table 1.

24

Flexibility to image

\*Determined by the sampling rate.

Geometrical invariance of

Geometrical invariance of feature descriptor

Feature localization

combined to a new feature in SURF by:

where A is the true warp matrix. The threshold is chosen as 5 pixels, i.e., a correspondence is identified as a mismatch if the transfer error is larger than 5 pixels in any image direction.

MFAR, also called 1-precision [10], is defined as:

$$MFAR = \frac{\text{\textquotedblleft matches\textquotedblright} - \text{\textquotedblleft correct\textquotedblright matches\textquotedblright}}{\text{\textquotedblleft matches\textquotedblleft}}\tag{17}$$

where "#" denotes "the number of." MFAR is just the rate of mismatches, which is related to image speckling as well as the radiometric and geometrical warping. It can be used together with #correct matches to evaluate the robustness of a detector to speckles on SAR image pair with controlled radiometric and geometrical warping.

WMEE is used to evaluate the consistency of the retrieved warp matrix and its true value:

$$\text{WMEE} = ||\mathbf{A} - \mathbf{A}\_{\mathbf{r}}||\_{F} \tag{18}$$

Detectors

27

 Estimated affine warping parameters

ab

True value

FH-15(-2)

FH-9(-1)

FH-9(-2)

FH-9(-3)

FH-9(-4)

FH-9(-5)

True value

FH-15(-2)

FH-9(-1)

FH-9(-2)

FH-9(-3)

FH-9(-4)

FH-9(-5)

True value

FH-15(-2)

FH-9(-1)

FH-9(-2)

FH-9(-3)

FH-9(-4)

FH-9(-5)

 1.1363

 0.1037

0.0894

 1.3160

2.4582

 5.5270

 1.1365

 0.1037

0.0894

 1.3159

2.4575

 5.5336

 1.1363

 0.1038

0.0895

 1.3165

2.4408

 5.4805

 1.1361

 0.1038

0.0897

 1.3160

2.3833

 5.6308

 1.1402

 0.1055

0.0805

 1.3160

 1.1387

 0.0984

0.0736

 1.3238

2.2175

3.6578

 3.5156

 1.7098

 1.1365

 0.1036

0.0894

 1.3159

 0.9360

 0.1890

0.1616

 1.0938

10.4227

2.6000

 5.4000

3.4476

 0.9361

 0.1890

0.1617

 1.0937

10.4252

3.4490

 0.9361

 0.1890

0.1613

 1.0937

10.4329

3.4817

 0.9352

 0.1898

0.1618

 1.0940

 0.9298

 0.1909

0.1603

 1.0868

9.9304

10.4452

3.4432

2.2059

 0.9370

 0.1887

0.1576

 1.0908

 0.9361

 0.1889

0.1617

 1.0938

 0.7186

 0.0457

0.0398

 0.8094

 1.4895 10.5000

10.5603

3.7546

55 (0.0678) 25 (0.2188) 170 (0.0449)

419 (0.0141)

735 (0.0252) 893 (0.0262)

—

47 (0.0408)

29 (0.1212)

157 (0.0427) 476 (0.0206) 983 (0.0180) 1293 (0.0300)

3.4000

 2.1085

 0.7181

 0.0453

0.0402

 0.8093

 1.6070

 2.1746

 0.7192

 0.0450

0.0403

 0.8088

 1.4752

 2.3425

 0.7186

 0.0458

0.0395

 0.8085

 1.3565

 2.2052

 0.7195

 0.0425

0.0347

 0.8067

 2.0444

 1.3480

 0.7164

 0.0415

0.0481

 0.8059

 2.3151

 3.6269

 0.7189

 0.0452

 c 0.0402

 0.8087

 1.7000

 2.4000

—

42 (0.1923)

22 (0.2414) 73 (0.1300) 129 (0.1164) 188 (0.1754) 176 (0.1619)

—

—

(0.7109, 0.8770) (0.5887, 0.7854) (0.6219, 0.6602) (0.3001, 0.4602) (0.2580, 0.3790) (0.2819, 0.3874)

—

(0.6040, 0.7075) (0.7949, 1.2405) (0.3821, 0.5200) (0.2267, 0.3080) (0.1703, 0.2143) (0.1601, 0.2273)

—

(0.7131, 0.9156) (1.0153, 0.9129) (0.3856, 0.4829) (0.1902, 0.3197) (0.1616, 0.2378) (0.1432, 0.2119)

 3.7101

 2.1610

 0.3166

 0.1784

 0.1954

 0.1903

 —

 0.3598

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

 1.3231

 0.0698

 0.1058

 0.0894

 0.0908

 —

 1.3725

 1.1070

 0.3949

DOI: http://dx.doi.org/10.5772/intechopen.81665

 0.2321

 0.2439

 0.3596

 —

 d

 tx

ty

Correct match number and MFAR

 ATE

WMEE

where k�k<sup>F</sup> denotes the Frobenius norm.

We evaluate the two SURF detectors on four image pairs with different transformations, the retrieved warp matrix parameters, ATE, correct match number, MFAR, and WMEE are listed in Table 2. It shows that FH-15(-2) can extract more correct matches with lower MFAR than FH-9(-1). This validates the robustness of SURF to speckling because FH-15(-2) performs the feature extraction on the double-sized image with much serious speckle. ATE of FH-15(-2) is smaller than that of FH-9(-1) except on the first image pair. On all the four pairs, the features extracted by FH-15(-2) can obtain subpixel estimation in both image directions, but FH-9(-1) obtains this only on the first pair. Therefore, FH-15(-2) features are more consistent with the retrieval parameters. This also signifies that FH-15(-2) can attains lower MFAR than FH-9(-1) because parameter estimation in EF-LTS is related to the outlier percentage in data. This will be detailed in Section 4. As on WMEE, the two detectors perform equally, FH-15(-2) does not improve the registration accuracy on all the four pairs as we expected, and there is still clear inconsistency between the retrieved warp matrix and the true value. The reason lies in that the sampling step is also doubled when FH-15(-2) doubles the image. This makes sampling being still conducted on the equivalently same pixel position rather than the subpixel image position. For instance, let (x0, y0) be a sampled pixel in the original image, the corresponding position in doubled image is (2x0, 2y0). The doubled step then makes this pixel position be still sampled instead of (2x<sup>0</sup> � 1, 2y<sup>0</sup> � 1), while the latter corresponds to the subpixel position (x<sup>0</sup> � 0.5, y<sup>0</sup> � 0.5) in the original image and positively contributes to the subpixel registration. Based on this, we suggest initializing SSP by using 9 � 9 box filter to the oversampled image but with unchanged sampling, we denote this detector as FH-9(-Fs), Fs denotes the sampling rate. To avoid nonlinear aliasing, the linear interpolator such as bilinear interpolator is used to conduct the sampling. Table 2 further summarizes the registration results based on FH-9(-2) to FH-9(-5) detector. Comparing with FH-9 (-1) and FH-15(-2), the correct match number, ATE, MFAR, and WMEE of FH-9 (-2) are all clearly improved. As oversampling rate increases from 2 to 5, the registration accuracy is also improved for more correspondences of higher localization accuracy are identified. All these make the high accurate SAR image registration possible. In view of the fact that oversampling will increase dataset and computational load, for high accuracy registration we recommend oversampling the image three or four times so as to achieve the compromise among accuracy, robustness, and computational complexity.


#### On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

where

where " #

true value:

2y 0

26

where k�k

pixels in any image direction.

" denotes

A is the true warp matrix. The threshold is chosen as 5 pixels, i.e., a

�

is related to image speckling as well as the radiometric and geometrical warping. It can be used together with #correct matches to evaluate the robustness of a detector to speckles on SAR image pair with controlled radiometric and geometrical warping. WMEE is used to evaluate the consistency of the retrieved warp matrix and its

¼ k k A � Ar

We evaluate the two SURF detectors on four image pairs with different transformations, the retrieved warp matrix parameters, ATE, correct match number, MFAR, and WMEE are listed in Table 2. It shows that FH-15(-2) can extract more correct matches with lower MFAR than FH-9(-1). This validates the robustness of SURF to speckling because FH-15(-2) performs the feature extraction on the double-sized image with much serious speckle. ATE of FH-15(-2) is smaller than that of FH-9(-1) except on the first image pair. On all the four pairs, the features extracted by FH-15(-2) can obtain subpixel estimation in both image directions, but FH-9(-1) obtains this only on the first pair. Therefore, FH-15(-2) features are more consistent with the retrieval parameters. This also signifies that FH-15(-2) can attains lower MFAR than FH-9(-1) because parameter estimation in EF-LTS is related to the outlier percentage in data. This will be detailed in Section 4. As on WMEE, the two detectors perform equally, FH-15(-2) does not improve the registration accuracy on all the four pairs as we expected, and there is still clear inconsistency between the retrieved warp matrix and the true value. The reason lies in that the sampling step is also doubled when FH-15(-2) doubles the image. This makes sampling being still conducted on the equivalently same pixel position rather

#correct matches

x 0 , y

#matches (17)

…

" MFAR is just the rate of mismatches, which

<sup>F</sup> (18)

0) be a sampled pixel in the

0). The

0 � 1,

� 0.5) in

0, 2y

x 0 � 0.5, y 0

� 9 box filter to the oversampled image

, MFAR, and WMEE of FH-9

correspondence is identified as a mismatch if the transfer error is larger than 5

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami

#matches

WMEE

MFAR, also called 1-precision [10], is defined as:

¼

"the number of.

<sup>F</sup> denotes the Frobenius norm.

than the subpixel image position. For instance, let (

(-1) and FH-15(-2), the correct match number, ATE

this, we suggest initializing SSP by using 9

ness, and computational complexity.

original image, the corresponding position in doubled image is (2x

� 1), while the latter corresponds to the subpixel position (

doubled step then makes this pixel position be still sampled instead of (2x

(-2) are all clearly improved. As oversampling rate increases from 2 to 5, the registration accuracy is also improved for more correspondences of higher localization accuracy are identified. All these make the high accurate SAR image registration possible. In view of the fact that oversampling will increase dataset and

the original image and positively contributes to the subpixel registration. Based on

but with unchanged sampling, we denote this detector as FH-9(-Fs), Fs denotes the sampling rate. To avoid nonlinear aliasing, the linear interpolator such as bilinear interpolator is used to conduct the sampling. Table 2 further summarizes the registration results based on FH-9(-2) to FH-9(-5) detector. Comparing with FH-9

computational load, for high accuracy registration we recommend oversampling the image three or four times so as to achieve the compromise among accuracy, robust-

MFAR


Table 2. Evaluation of SURF Fast-Hessian detectors on four controlled SAR image pairs. 4. Appropriate retrieval algorithm for SAR image registration

DOI: http://dx.doi.org/10.5772/intechopen.81665

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

the attained correspondences. Due to the influences of spatial/temporal

spondences, some robust outlier-insensitive algorithms are necessary.

4.1 Evaluation of RANSAC for SAR image registration

local discontentment.

any MSS of inliers can generate the

29

The next procedure after feature extraction is to retrieve the warp function from

decorrelation, system noise, and environmental interference, or the non-robustness in the depiction and matching of features, there are always mismatches in the constructed correspondences. It is difficult to get a priori information to remove them beforehand. To accurately retrieve parameters from these error-prone corre-

Furthermore, unlike the pinhole imaging of optical camera, SAR acquires the imagery using a slant-range geometry which cannot be modeled as a central projection [47]. As a result, the warp model between SAR images is dependent on the system parameter, imaging geometry, and target relief, and we cannot adopt a global homography or essential matrix to model the geometrical warping then. Nevertheless, when the system parameter and imaging geometry are fixed and the area-of-interest has gentle topography, we can conventionally approximate the warp function as a low-order polynomial [48]. This indicates our strategy in the retrieval of registration parameters, to focus on the global registration instead of

RANSAC [30] has been widely used in feature-based SAR image registrations for parameter retrieval [15, 16, 26, 27]. Unlike LS which uses all the available data to estimate parameters, RANSAC conducts the estimation using a few-to-many strategy or a local-to-global strategy. A MSS is randomly sampled from the constructed correspondences to achieve an estimation of the warp function firstly. The cardinality of MSS, i.e., the smallest sufficiency to determine the warp parameters, is just related to the degree of freedom (DoF) of the warp function. For example, the cardinality will be 3 for affine transformation of 6 DOFs. The entire dataset are then checked for those correspondences consistent with the retrieved warping to construct a larger CS. These two steps are repeated until the largest CS is finally achieved for parameter estimation. This local-to-global strategy is tenable only if

"true value

often hard to keep this in real registration due to the unavoidable noise and local distortion, i.e., a different estimation of parameters will be achieved from a different MSS configuration of inliers. This uncertainty is even more severe in SAR image registration because SAR warping varies from pixel to pixel and the low-order polynomial approximation only accounts for global registration instead of local contentment. The local-to-global strategy may then magnify the local distortion, aggravate the estimation uncertainty, and damnify the global registration accuracy although a largest CS is identified. To demonstrate this, we devise an experiment to coregister a spaceborne InSAR image pair as shown in Figure 4(a) and (b). The two images are acquired by RadarSat-2 on May 4 and 28, 2008, respectively. The scene is within South Phoenix, AZ, USA with some buildings and vegetable lands. We first use FH-9(-1) to construct SURF feature correspondences, and then adopt RANSAC to retrieve the affine warp parameters. To evaluate the estimation certainty, we execute RANSAC 100 times and based on the obtained parameters of each execution, we coregister the complex image pair to calculate the three-look coherent CC and spectral SNR. CC measures the consistency, while spectral SNR, the ratio between the maximum entry and the sum of other entries in the spectrum, reflects the clarity of the interferogram fringe [49]. Figure 5 displays the affine

" of warp parameters [31]. But it is

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

## 4. Appropriate retrieval algorithm for SAR image registration

The next procedure after feature extraction is to retrieve the warp function from the attained correspondences. Due to the influences of spatial/temporal decorrelation, system noise, and environmental interference, or the non-robustness in the depiction and matching of features, there are always mismatches in the constructed correspondences. It is difficult to get a priori information to remove them beforehand. To accurately retrieve parameters from these error-prone correspondences, some robust outlier-insensitive algorithms are necessary.

Furthermore, unlike the pinhole imaging of optical camera, SAR acquires the imagery using a slant-range geometry which cannot be modeled as a central projection [47]. As a result, the warp model between SAR images is dependent on the system parameter, imaging geometry, and target relief, and we cannot adopt a global homography or essential matrix to model the geometrical warping then. Nevertheless, when the system parameter and imaging geometry are fixed and the area-of-interest has gentle topography, we can conventionally approximate the warp function as a low-order polynomial [48]. This indicates our strategy in the retrieval of registration parameters, to focus on the global registration instead of local discontentment.

#### 4.1 Evaluation of RANSAC for SAR image registration

RANSAC [30] has been widely used in feature-based SAR image registrations for parameter retrieval [15, 16, 26, 27]. Unlike LS which uses all the available data to estimate parameters, RANSAC conducts the estimation using a few-to-many strategy or a local-to-global strategy. A MSS is randomly sampled from the constructed correspondences to achieve an estimation of the warp function firstly. The cardinality of MSS, i.e., the smallest sufficiency to determine the warp parameters, is just related to the degree of freedom (DoF) of the warp function. For example, the cardinality will be 3 for affine transformation of 6 DOFs. The entire dataset are then checked for those correspondences consistent with the retrieved warping to construct a larger CS. These two steps are repeated until the largest CS is finally achieved for parameter estimation. This local-to-global strategy is tenable only if any MSS of inliers can generate the "true value" of warp parameters [31]. But it is often hard to keep this in real registration due to the unavoidable noise and local distortion, i.e., a different estimation of parameters will be achieved from a different MSS configuration of inliers. This uncertainty is even more severe in SAR image registration because SAR warping varies from pixel to pixel and the low-order polynomial approximation only accounts for global registration instead of local contentment. The local-to-global strategy may then magnify the local distortion, aggravate the estimation uncertainty, and damnify the global registration accuracy although a largest CS is identified. To demonstrate this, we devise an experiment to coregister a spaceborne InSAR image pair as shown in Figure 4(a) and (b). The two images are acquired by RadarSat-2 on May 4 and 28, 2008, respectively. The scene is within South Phoenix, AZ, USA with some buildings and vegetable lands. We first use FH-9(-1) to construct SURF feature correspondences, and then adopt RANSAC to retrieve the affine warp parameters. To evaluate the estimation certainty, we execute RANSAC 100 times and based on the obtained parameters of each execution, we coregister the complex image pair to calculate the three-look coherent CC and spectral SNR. CC measures the consistency, while spectral SNR, the ratio between the maximum entry and the sum of other entries in the spectrum, reflects the clarity of the interferogram fringe [49]. Figure 5 displays the affine

Detectors

28

 Estimated affine warping parameters

ab

True value

FH-15(-2)

FH-9(-1)

FH-9(-2)

FH-9(-3)

FH-9(-4)

FH-9(-5)

Table 2. Evaluation

 of SURF

Fast-Hessian

 detectors on four controlled SAR image pairs.

 1.2077

 0.0778

0.0719

 1.3078

5.1297

 1.6397

 1.2078

 0.0777

0.0718

 1.3077

5.1181

 1.6590

 1.2076

 0.0766

0.0719

 1.3077

5.0751

 1.6740

 1.2075

 0.0778

0.0735

 1.3066

5.0414

 2.0074

 1.1959

 0.0695

0.0650

 1.3079

1.8954

 0.0939

 1.2033

 0.0744

0.0753

 1.3054

4.2454

 2.4055

 1.2079

 0.0777

 c 0.0718

 1.3077

 d

 tx 5.3000

 1.5000

—

52 (0.0545)

24 (0.0769) 172 (0.0227) 514 (0.0172)

1052 (0.0177)

1451 (0.0176)

—

(0.7247, 0.7616) (0.9570, 1.1486) (0.3858, 0.4182) (0.2207, 0.3552) (0.1506, 0.2289) (0.1343, 0.1998)

 1.3900

 3.6836

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

 0.5695

 0.2844

 0.2416

 0.2203

 —

ty

Correct match number and MFAR

 ATE

WMEE

## Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

parameters a, b, c, d, tx, and ty as well as CC and SNR obtained in each execution. Table 3 further displays the mean and standard deviation of the parameters, CC, and SNR. RANSAC cannot obtain a stable registration because the retrieval parameters vary with executions, even for executions with the same cardinality of CS achieved. Figure 6 shows the retrieval parameters, CC and SNR for 48 executions with the same cardinality. We can still find the estimation uncertainty. This reveals that the attained inliers which compose the final CS are actually different although the same cardinality. Otherwise, the parameters would be the same for each execution because they are retrieved by just LS fitting the inliers.

The uncertainty of RANSAC in SAR image registration just comes from its retrieval strategy and loss function. To achieve a stable registration for SAR images, a feasible improvement is to estimate the parameters with more correspondences to reflect the true support than just a MSS, and to apply an appropriate loss function. This leads us another direction to the robust parameter regression.

#### Figure 4.

Registration of InSAR image pair from RadarSat-2. (a) Master image, (b) slave image, and the final (c) interferogram and (d) correlation map based on the registration parameters estimated by EF-LTS.

#### Figure 5.

Registration parameters, cross-correlation (CC), and spectral SNR obtained by 100 executions of (thin line) RANSAC and (thick line) EF-LTS on the image pair of RadarSat-2.

Algorithm

31

a

> Mean

> > RANSAC

EF-LTS Algorithm

0.9990

0.0000

d

Mean

RANSAC

EF-LTS

Table 3.

Mean and standard deviation of the registration

 parameters,

cross-correlation

 (CC), and spectral SNR obtained by RANSAC and EF-LTS on RadarSat-2

0.9996

0.0000

 0.9996

 2.2264 104

Std

Mean 2.6068 2.6151

Std 0.1955 0.0000

Mean 0.5133 0.5008

Std 0.1509

0.0000

37.26

 InSAR images.

 0.0000

37.36

 0.1126

Mean

Std

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

2.5485 104

 0.9992

 2.5228 104

Std

Mean

1.8796 104

2.4879 104

0.0000

> tx

9.2440 105

1.3770 104

3.3408 104

0.0000

> ty

0.5483

 0.0000

> SNR (dB)

Std

Mean

Std

Mean 0.5462

 0.0046

Std

b

c

CC

DOI: http://dx.doi.org/10.5772/intechopen.81665


On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

parameters a, b, c, d, tx, and ty as well as CC and SNR obtained in each execution. Table 3 further displays the mean and standard deviation of the parameters, CC, and SNR. RANSAC cannot obtain a stable registration because the retrieval parameters vary with executions, even for executions with the same cardinality of CS achieved. Figure 6 shows the retrieval parameters, CC and SNR for 48 executions with the same cardinality. We can still find the estimation uncertainty. This reveals that the attained inliers which compose the final CS are actually different although the same cardinality. Otherwise, the parameters would be the same for each execu-

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

The uncertainty of RANSAC in SAR image registration just comes from its retrieval strategy and loss function. To achieve a stable registration for SAR images, a feasible improvement is to estimate the parameters with more correspondences to reflect the true support than just a MSS, and to apply an appropriate loss function.

Registration of InSAR image pair from RadarSat-2. (a) Master image, (b) slave image, and the final (c) interferogram and (d) correlation map based on the registration parameters estimated by EF-LTS.

Registration parameters, cross-correlation (CC), and spectral SNR obtained by 100 executions of (thin line)

RANSAC and (thick line) EF-LTS on the image pair of RadarSat-2.

tion because they are retrieved by just LS fitting the inliers.

Figure 4.

Figure 5.

30

This leads us another direction to the robust parameter regression.

Algorithm 1: C-step

)π(1) ≤ ��� ≤ (rold

Algorithm 2: Fast-LTS

)π(1) ≤ ��� ≤ (r0

when n is larger [33].

uals rold

(rold 2

(r0 2

tion θ.

Step 11. Compute regression parameters θold by LS fitting Hold.

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

<sup>2</sup> for a permutation π of the set such that

)π(h) ≤ ��� ≤ (r0

xsi ¼ ∑ N j¼0 ∑ N�j k¼0 ajkx j miy<sup>k</sup> mi þ ς<sup>i</sup>

ysi ¼ ∑ N j¼0 ∑ N�j k¼0 bjkx j miy<sup>k</sup> mi þ ξ<sup>i</sup>

θ ¼ θ1; θ2; ⋯; θ<sup>p</sup>

ψ ¼ ψ1; ψ2; ⋯; ψ<sup>p</sup> h i<sup>T</sup>

)π(h) ≤ ��� ≤ (rold

a few C-steps. Thus Fast-LTS conducts estimation as follows [33]:

2

2

new parameters θnew by LS fitting Hnew.

DOI: http://dx.doi.org/10.5772/intechopen.81665

2

H0. Repeat above procedures 500 times.

4.3 EF-LTS for SAR image registration

8 >>>><

>>>>:

2D linear regression problem:

with

8 >>>><

>>>>:

<sup>i</sup> θ þ ς<sup>i</sup>

<sup>i</sup> ψ þ ξ<sup>i</sup>

xsi <sup>¼</sup> <sup>X</sup><sup>T</sup>

8 >>>><

>>>>:

33

ysi <sup>¼</sup> <sup>X</sup><sup>T</sup>

Step 12. Calculate residuals rold based on θold. Ascendingly sort squared resid-

2 )π(<sup>n</sup>). Step 13. Construct a new h-subset Hnew = {π(1), π(2), …, π(h)} and obtain the

It has been proved that Q of parameters θnew is always no larger than that of parameters θold [33]. Therefore, an improved estimation of parameters can be achieved after an execution of C-step, and a converged Q will be obtained after only

Step 21. Randomly generate a p-subset as parameter set θ0. Calculate n residuals r0 based on θ<sup>0</sup> to achieve an initial h-subset H0 = {π(1), π(2), …, π(h)} such that

Step 22. Implement C-steps on the 10 H0 with the lowest 10 Q until convergence. Then the solution that creates the lowest Q is identified as the final estima-

The trimming constant h is set between [(n+p+ 1)/2) ([x) denotes the smallest integer larger than x) and n. The breakdown value of Fast-LTS is (n � h + 1)/n. A nested extension approach should be adopted to enable an efficient estimation

Fast-LTS is appropriate for 1D linear regression formulated in (19). However, for SAR image registration, what we need to do is to fit a 2D polynomial regression

where n is the number of constructed correspondences, N is the order of poly-

� �<sup>T</sup> <sup>¼</sup> ½ � <sup>a</sup>00; <sup>a</sup>01; <sup>⋯</sup>; aN<sup>0</sup> <sup>T</sup>

<sup>X</sup><sup>i</sup> <sup>¼</sup> Xi1; Xi2; <sup>⋯</sup>;Xip � �<sup>T</sup> <sup>¼</sup> <sup>1</sup>; ymi; <sup>⋯</sup>; <sup>y</sup><sup>N</sup>

where θ and ψ are the unknown parameters to be estimated, and p = (N + 1) (N + 2)/2 denotes the number of unknowns. Then, the warp function estimation for SAR image registration can be transformed into the following optimization problems:

<sup>¼</sup> ½ � <sup>b</sup>00; <sup>b</sup>01; <sup>⋯</sup>; bN<sup>0</sup> <sup>T</sup>

mi; xmi; ⋯; x<sup>N</sup> mi � �<sup>T</sup>

nomial, a and b are polynomial coefficients, (xsi, ysi) and (xmi, ymi) are the ith feature correspondence extracted from the slave and master images, and ζ<sup>i</sup> and ξ<sup>i</sup> denote the normally distributed error terms with zero mean. Actually, (21) denote a

)π(<sup>n</sup>). Update H0 by carrying out two C-steps on

, i ¼ 1, …, n

(21)

, i ¼ 1, …, n

(22)

Figure 6.

Registration parameters, cross-correlation (CC), and spectral SNR obtained by 48 executions of RANSAC on RadarSat-2 InSAR images with the same CS cardinality.

#### 4.2 Fast-LTS

The widely used LS is now being criticized more and more for lack of robustness. To tackle with this, some robust regression approaches were developed, like LMedS [32] and the least trimmed squares (LTS) [50]. LMedS implements the regression by minimizing the median of residual squares. This makes LMedS so robust that it can still obtain a reasonable estimation even if 50% of the dataset are outliers. So the breakdown point of LMedS is as high as 50%. LTS is a modification of LS with the same breakpoint as LMedS. It also fits the linear model:

$$y\_i = \mathbf{X}\_i^\mathrm{T} \mathbf{\hat{o}} + e\_i, \ i = \mathbf{1}, \ldots, n \tag{19}$$

where X<sup>i</sup> = [xi1, xi2, …, xip] <sup>T</sup> denotes the explanatory variable, yi denotes the response variable, θ = [θ1, θ2, …, θp] <sup>T</sup> indicates the unknown parameter to be retrieved, ei is the error term, n is the sample size, and p is the dimension of Xi. The loss function of LTS is:

$$Q \coloneqq \text{Minimize } \sum\_{i=1}^{h} \left(\mathbf{r}^2\right)\_i \text{ with } \mathbf{r} = \left[r\_1, r\_2, \dots, r\_n\right]^\mathrm{T} \text{ and } r\_i = y\_i - \mathbf{X}\_i^\mathrm{T} \mathbf{\theta} \tag{20}$$

where (r 2 )<sup>i</sup> denotes the ith element of the ordered squared residuals (r 2 )1 ≤ ��� ≤ (r 2 )<sup>i</sup> ≤ ��� ≤ (r 2 )n, and h is termed as the trimming constant. LTS conducts regression by LS fitting the h-subset to minimize the squared residuals. Compared with LMedS, the statistical efficiency of LTS is much better and the loss function is much smoother [33]. Nevertheless, the deficiency of LTS is the large computation when processing the big data. To accelerate it, Rousseeuw and Van Driessen [33] developed a Fast-LTS, which can efficiently deal with a sample size as large as tens of thousands or even larger. The core of Fast-LTS is a concentration step (C-step), which is designed to achieve a better estimation from an old h-subset Hold [33]:

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

#### Algorithm 1: C-step

Step 11. Compute regression parameters θold by LS fitting Hold.

Step 12. Calculate residuals rold based on θold. Ascendingly sort squared residuals rold <sup>2</sup> for a permutation π of the set such that

(rold 2 )π(1) ≤ ��� ≤ (rold 2 )π(h) ≤ ��� ≤ (rold 2 )π(<sup>n</sup>).

Step 13. Construct a new h-subset Hnew = {π(1), π(2), …, π(h)} and obtain the new parameters θnew by LS fitting Hnew.

It has been proved that Q of parameters θnew is always no larger than that of parameters θold [33]. Therefore, an improved estimation of parameters can be achieved after an execution of C-step, and a converged Q will be obtained after only a few C-steps. Thus Fast-LTS conducts estimation as follows [33]:

#### Algorithm 2: Fast-LTS

Step 21. Randomly generate a p-subset as parameter set θ0. Calculate n residuals r0 based on θ<sup>0</sup> to achieve an initial h-subset H0 = {π(1), π(2), …, π(h)} such that (r0 2 )π(1) ≤ ��� ≤ (r0 2 )π(h) ≤ ��� ≤ (r0 2 )π(<sup>n</sup>). Update H0 by carrying out two C-steps on H0. Repeat above procedures 500 times.

Step 22. Implement C-steps on the 10 H0 with the lowest 10 Q until convergence. Then the solution that creates the lowest Q is identified as the final estimation θ.

The trimming constant h is set between [(n+p+ 1)/2) ([x) denotes the smallest integer larger than x) and n. The breakdown value of Fast-LTS is (n � h + 1)/n. A nested extension approach should be adopted to enable an efficient estimation when n is larger [33].

#### 4.3 EF-LTS for SAR image registration

Fast-LTS is appropriate for 1D linear regression formulated in (19). However, for SAR image registration, what we need to do is to fit a 2D polynomial regression

$$\begin{cases} \mathbf{x}\_{si} = \sum\_{j=0}^{N} \sum\_{k=0}^{N-j} a\_{jk} \mathbf{x}\_{mi}^{j} y\_{mi}^{k} + \boldsymbol{\varepsilon}\_{i} \\\\ \mathbf{y}\_{si} = \sum\_{j=0}^{N} \sum\_{k=0}^{N-j} b\_{jk} \mathbf{x}\_{mi}^{j} y\_{mi}^{k} + \boldsymbol{\xi}\_{i} \end{cases}, \ i = 1, \ldots, n \tag{21}$$

where n is the number of constructed correspondences, N is the order of polynomial, a and b are polynomial coefficients, (xsi, ysi) and (xmi, ymi) are the ith feature correspondence extracted from the slave and master images, and ζ<sup>i</sup> and ξ<sup>i</sup> denote the normally distributed error terms with zero mean. Actually, (21) denote a 2D linear regression problem:

$$\begin{cases} \mathbf{x}\_{ii} = \mathbf{X}\_{i}^{\mathsf{T}} \boldsymbol{\Phi} + \boldsymbol{\varepsilon}\_{i} \\ \boldsymbol{\nu}\_{ii} = \mathbf{X}\_{i}^{\mathsf{T}} \boldsymbol{\Psi} + \boldsymbol{\xi}\_{i} \end{cases} \text{ with } \begin{cases} \boldsymbol{\Phi} = \begin{bmatrix} \boldsymbol{\theta}\_{1}, \boldsymbol{\theta}\_{2}, \cdots, \boldsymbol{\theta}\_{p} \end{bmatrix}^{\mathsf{T}} = \begin{bmatrix} a\_{00}, a\_{01}, \cdots, a\_{N0} \end{bmatrix}^{\mathsf{T}} \\\ \boldsymbol{\Psi} = \begin{bmatrix} \boldsymbol{\nu}\_{1}, \boldsymbol{\nu}\_{2}, \cdots, \boldsymbol{\nu}\_{p} \end{bmatrix}^{\mathsf{T}} = \begin{bmatrix} b\_{00}, b\_{01}, \cdots, b\_{N0} \end{bmatrix}^{\mathsf{T}} \\\ \mathbf{X}\_{i} = \begin{bmatrix} \boldsymbol{X}\_{i1}, \boldsymbol{X}\_{i2}, \cdots, \boldsymbol{\mathcal{X}}\_{ip} \end{bmatrix}^{\mathsf{T}} = \begin{bmatrix} 1, y\_{mi}, \cdots, y\_{mi}^{N}, \boldsymbol{x}\_{mi}, \cdots, \boldsymbol{x}\_{mi}^{N} \end{bmatrix}^{\mathsf{T}} \end{cases} , \quad i = 1, \ldots, n \end{cases} \tag{22}$$

where θ and ψ are the unknown parameters to be estimated, and p = (N + 1) (N + 2)/2 denotes the number of unknowns. Then, the warp function estimation for SAR image registration can be transformed into the following optimization problems:

4.2 Fast-LTS

Figure 6.

The widely used LS is now being criticized more and more for lack of robustness. To tackle with this, some robust regression approaches were developed, like LMedS [32] and the least trimmed squares (LTS) [50]. LMedS implements the regression by minimizing the median of residual squares. This makes LMedS so robust that it can still obtain a reasonable estimation even if 50% of the dataset are outliers. So the breakdown point of LMedS is as high as 50%. LTS is a modification of LS with the

Registration parameters, cross-correlation (CC), and spectral SNR obtained by 48 executions of RANSAC on

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

retrieved, ei is the error term, n is the sample size, and p is the dimension of Xi. The

)<sup>i</sup> denotes the ith element of the ordered squared residuals

conducts regression by LS fitting the h-subset to minimize the squared residuals. Compared with LMedS, the statistical efficiency of LTS is much better and the loss function is much smoother [33]. Nevertheless, the deficiency of LTS is the large computation when processing the big data. To accelerate it, Rousseeuw and Van Driessen [33] developed a Fast-LTS, which can efficiently deal with a sample size as large as tens of thousands or even larger. The core of Fast-LTS is a concentration step (C-step), which is designed to achieve a better estimation from an old h-subset

<sup>i</sup> with <sup>r</sup> <sup>¼</sup> ½ � <sup>r</sup>1;r2; <sup>⋯</sup>;rn <sup>T</sup> and ri <sup>¼</sup> yi � <sup>X</sup><sup>T</sup>

)n, and h is termed as the trimming constant. LTS

<sup>i</sup> θ þ ei, i ¼ 1, …, n (19)

<sup>i</sup> θ (20)

<sup>T</sup> denotes the explanatory variable, yi denotes the

<sup>T</sup> indicates the unknown parameter to be

same breakpoint as LMedS. It also fits the linear model:

where X<sup>i</sup> = [xi1, xi2, …, xip]

loss function of LTS is:

where (r

Hold [33]:

32

)1 ≤ ��� ≤ (r

(r 2 Q≔Minimize ∑

2

2

response variable, θ = [θ1, θ2, …, θp]

h i¼1 r<sup>2</sup>

RadarSat-2 InSAR images with the same CS cardinality.

)<sup>i</sup> ≤ ��� ≤ (r

2

yi <sup>¼</sup> <sup>X</sup><sup>T</sup>

$$\begin{cases} Q\_{\mathbf{x}} \coloneqq \text{Minimize } \sum\_{i=1}^{h} \left( \mathbf{r}\_{\mathbf{x}}^{2} \right)\_{i} \\\ Q\_{\mathbf{y}} \coloneqq \text{Minimize } \sum\_{i=1}^{h} \left( \mathbf{r}\_{\mathbf{y}}^{2} \right)\_{i} \end{cases} \text{with} \begin{cases} \mathbf{r}\_{\mathbf{x}} = \left[ r\_{\mathbf{x}1}, r\_{\mathbf{x}2}, \dots, r\_{\mathbf{x}n} \right]^{\mathrm{T}} \\\ \mathbf{r}\_{\mathbf{y}} = \left[ r\_{j1}, r\_{j2}, \dots, r\_{jn} \right]^{\mathrm{T}} \end{cases} \text{and} \begin{cases} r\_{\mathbf{x}i} = \mathbf{x}\_{i\mathbf{i}} - \mathbf{X}\_{i}^{\mathrm{T}} \mathbf{0} \\\ r\_{ji} = \mathbf{y}\_{si} - \mathbf{X}\_{i}^{\mathrm{T}} \mathbf{W} \end{cases}, \ i = 1, \dots, n \tag{23}$$

wxi <sup>¼</sup> 1 if j j rxi=σ<sup>x</sup> <sup>≤</sup>2:<sup>5</sup>

0 if j j rxi=σ<sup>x</sup> > 2:5

DOI: http://dx.doi.org/10.5772/intechopen.81665

( (

attained by LS solving the following optimizations:

which in fact indicates the weighted LS.

residuals larger than 2.5σ in a Gaussian situation [50].

of inliers, a good estimation of q can be obtained by

with smaller q, as listed in Table 4.

35

8 >><

>>:

wyi <sup>¼</sup> 1 if ryi=σ<sup>y</sup>

where "&" denotes the logical AND operator. The final estimations θ<sup>f</sup> and ψ<sup>f</sup> are

n i¼1 wir 2 xi

n i¼1 wir 2 yi

θ<sup>f</sup> ¼ argmin∑

ψ<sup>f</sup> ¼ argmin∑

Step 33 makes EF-LTS obtain more accurate and stable estimation than the original LTS. The logical AND in (29) shows that only the feature correspondence which is correctly matched in both x- and y-direction is considered as an inlier. This is necessary for accurate estimation because mismatching in one direction may also affect the matching in another. The bound in (28) is set as 2.5 for there are very few

In Fast-LTS, the random sampling number T is a constant 500. This is inappropriate because accurate estimation only requires one p-match to being "clean." Let q denote the percentage of inliers in data, then the probability ε of having at least one

Since the trimming constant h is chosen beforehand according to the percentage

<sup>q</sup>^ <sup>¼</sup> <sup>h</sup>

Therefore, if a required false alarm rate ε for the estimation is given, the sam-

Thus, iteration in EF-LTS is controlled by the inlier percentage rather than the inlier number. Table 4 shows the sampling number T under given N and q when ε = 0.99. It can be seen that even the worst sampling number 293 is much smaller than 500 for N = 2. Thus, the constant 500 sampling will be redundant for the second-order polynomial, but will be insufficient for the third-order polynomial

Thus, besides introducing more iterations and computation load, higher MFAR

will also lead to a smaller h-subset, which indicates more localization and less

n � � � �<sup>p</sup> " !

<sup>T</sup> <sup>¼</sup> log 1ð Þ � <sup>ε</sup> log 1 � <sup>h</sup>

"clean" p-match among all the T random p-matches can be expressed as

pling number T can be then calculated by combining (31) and (32):

The inlier percentage q is in fact related to MFAR by:

The credible correspondence in both directions of x and y is chosen:

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

� � �

0 if ryi=σ<sup>y</sup> � � �

�≤2:5

� > 2:5

wi ¼ wxi & wyi, i ¼ 1, …, n (29)

<sup>ε</sup> <sup>¼</sup> <sup>1</sup> � <sup>1</sup> � <sup>q</sup><sup>p</sup> ð Þ<sup>T</sup>: (31)

q ¼ 1 � MFAR: (34)

<sup>n</sup> : (32)

: (33)

, i ¼ 1, …, n:

(28)

(30)

where (rx 2 )<sup>i</sup> represents the ith element of the ordered squared residuals (rx 2 )1 ≤ ��� ≤ (rx 2 )<sup>i</sup> ≤ ��� ≤ (rx 2 )n, and the meaning of (ry 2 )<sup>i</sup> can be likewise inferred. Each of the two optimizations in (23) is of the standard form (20). A direct solution to (23) may be thus achieved by decomposing 2D regression as two independent 1D regressions and using Fast-LTS to conduct estimation, respectively. This idea is feasible, but it may result in unnecessary computations because the feature positions in two image directions are in fact tied to each other, i.e., for the ith feature (xi, yi), the selection of xi will naturally mean the selection of yi. We can thus combine the two 1D regressions into a real 2D regression effectively, i.e., the extended Fast-LTS (EF-LTS):

#### Algorithm 3: EF-LTS

Step 31. Randomly draw p feature matches and LS fit them to estimate the initial parameters θ<sup>0</sup> and ψ0, and calculate the initial residuals r0x and r0y by

$$\begin{cases} \mathbf{r\_{0x}} = \begin{bmatrix} r\_{0x1}, r\_{0x2}, \dots, r\_{0xn} \end{bmatrix}^{\mathrm{T}} \text{ and } \begin{cases} r\_{0xi} = \mathbf{x}\_{si} - \mathbf{X}\_i^{\mathrm{T}} \mathbf{\theta\_0} \\ r\_{0yi} = \mathbf{y}\_{si} - \mathbf{X}\_i^{\mathrm{T}} \mathbf{\uprho\_0} \end{cases}, i = \mathbf{1}, \dots, n. \end{cases} \tag{24}$$

Then construct the initial h-subsets Hx0 and Hy0 by:

$$\begin{cases} \mathbf{Hx}\_{\mathbf{0}} = \{\mathfrak{x}\mathbf{x}(1), \mathfrak{x}\mathbf{x}(2), \dots, \mathfrak{x}\mathbf{x}(h)\} \subset \{1, 2, \dots, n\}\_{\mathbf{x}\mathbf{1}, \mathbf{f}}\\ \mathbf{Hy}\_{\mathbf{0}} = \{\mathfrak{x}\mathbf{y}(1), \mathfrak{x}\mathbf{y}(2), \dots, \mathfrak{x}\mathbf{y}(h)\} \subset \{1, 2, \dots, n\}^{\mathbf{f}, \mathbf{f}} \left( \begin{pmatrix} \mathbf{r}\_{\mathbf{0x}}^{2} \end{pmatrix}\_{\mathbf{x}\mathbf{y}(1)} \le \dots \le \left( \mathbf{r}\_{\mathbf{0x}}^{2} \right)\_{\mathbf{x}\mathbf{x}(n)} \right) \\ \left( \mathbf{r}\_{\mathbf{0y}}^{2} \right)\_{\mathbf{x}\mathbf{y}(1)} \le \dots \le \left( \mathbf{r}\_{\mathbf{0y}}^{2} \right)\_{\mathbf{x}\mathbf{y}(h)} \le \dots \le \left( \mathbf{r}\_{\mathbf{0y}}^{2} \right)\_{\mathbf{x}\mathbf{y}(n)} \end{cases} \tag{25}$$

Carry out two C-steps on Hx0 and Hy0 to obtain the h-subsets Hx2 and Hy2 with smaller Qx and Qy, respectively. Iteratively repeat above procedures T times to obtain a set of h-subsets Hx2 and Hy2.

Step 32. Select 10 Hx2 with the smallest 10 Qx and 10 Hy2 with the smallest 10 Qy if T is larger than 10; otherwise, select all Hx2 and Hy2. Carry out C-steps on these h-subsets until convergence. The solutions corresponding to the smallest Qx and Qy are selected as the raw estimations θ<sup>r</sup> and ψr, respectively.

Step 33. Calculate residuals rrx and rry based on θ<sup>r</sup> and ψr,

$$\begin{cases} \mathbf{r\_{rx}} = \left[r\_{rx1}, r\_{rx2}, \dots, r\_{rxn}\right]^\mathrm{T} \\ \mathbf{r\_{ry}} = \left[r\_{\gamma 1}, r\_{\gamma 2}, \dots, r\_{\gamma n}\right]^\mathrm{T} \end{cases} \text{with} \begin{cases} r\_{rxi} = \mathbf{x}\_{si} - \mathbf{X}\_i^\mathrm{T} \mathbf{\theta\_r} \\ r\_{\gamma i} = \mathbf{y}\_{si} - \mathbf{X}\_i^\mathrm{T} \mathbf{\uprho\_r} \end{cases}, i = \mathbf{1}, \dots, n \tag{26}$$

and estimate the error scales σ<sup>x</sup> and σ<sup>y</sup> by

$$\begin{cases} \sigma\_{\mathbf{x}} = \mathbf{C}\_{1} \sqrt{\frac{1}{h} \sum\_{i=1}^{h} (\mathbf{r}\_{\mathbf{rx}}^{2})\_{i}}\\\\ \sigma\_{\mathbf{y}} = \mathbf{C}\_{2} \sqrt{\frac{1}{h} \sum\_{i=1}^{h} \left(\mathbf{r}\_{\mathbf{ry}}^{2}\right)\_{i}} \end{cases} \tag{27}$$

where C<sup>1</sup> and C<sup>2</sup> are correction factors to achieve consistency at Gaussian error distributions [50]. Based on (27), we further calculate two weights by:

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

$$w\_{\rm xi} = \begin{cases} 1 & \text{if } \left| r\_{\rm xi} / \sigma\_{\rm x} \right| \le 2.5\\ 0 & \text{if } \left| r\_{\rm xi} / \sigma\_{\rm x} \right| > 2.5 \end{cases} \\ w\_{\rm yi} = \begin{cases} 1 & \text{if } \left| r\_{\rm ji} / \sigma\_{\rm y} \right| \le 2.5\\ 0 & \text{if } \left| r\_{\rm yi} / \sigma\_{\rm y} \right| > 2.5 \end{cases}, \\ i = 1, \ldots, n. \tag{28}$$

The credible correspondence in both directions of x and y is chosen:

$$w\_i = w\_{xi} \circledast w\_{yi}, \ i = 1, \ldots, n \tag{29}$$

where "&" denotes the logical AND operator. The final estimations θ<sup>f</sup> and ψ<sup>f</sup> are attained by LS solving the following optimizations:

$$\begin{cases} \boldsymbol{\Phi\_{\mathbf{f}}} = \operatorname\*{argmin} \sum\_{i=1}^{n} w\_{i} r\_{xi}^{2} \\\\ \boldsymbol{\Psi\_{\mathbf{f}}} = \operatorname\*{argmin} \sum\_{i=1}^{n} w\_{i} r\_{yi}^{2} \end{cases} \tag{30}$$

which in fact indicates the weighted LS.

Step 33 makes EF-LTS obtain more accurate and stable estimation than the original LTS. The logical AND in (29) shows that only the feature correspondence which is correctly matched in both x- and y-direction is considered as an inlier. This is necessary for accurate estimation because mismatching in one direction may also affect the matching in another. The bound in (28) is set as 2.5 for there are very few residuals larger than 2.5σ in a Gaussian situation [50].

In Fast-LTS, the random sampling number T is a constant 500. This is inappropriate because accurate estimation only requires one p-match to being "clean." Let q denote the percentage of inliers in data, then the probability ε of having at least one "clean" p-match among all the T random p-matches can be expressed as

$$\boldsymbol{\varepsilon} = \mathbf{1} - \left(\mathbf{1} - \mathbf{q}^{p}\right)^{T}.\tag{31}$$

Since the trimming constant h is chosen beforehand according to the percentage of inliers, a good estimation of q can be obtained by

$$
\hat{q} = \frac{h}{n}.\tag{32}
$$

Therefore, if a required false alarm rate ε for the estimation is given, the sampling number T can be then calculated by combining (31) and (32):

$$T = \left[\frac{\log\left(1 - \varepsilon\right)}{\log\left(1 - \left(\frac{h}{n}\right)^p\right)}\right].\tag{33}$$

Thus, iteration in EF-LTS is controlled by the inlier percentage rather than the inlier number. Table 4 shows the sampling number T under given N and q when ε = 0.99. It can be seen that even the worst sampling number 293 is much smaller than 500 for N = 2. Thus, the constant 500 sampling will be redundant for the second-order polynomial, but will be insufficient for the third-order polynomial with smaller q, as listed in Table 4.

The inlier percentage q is in fact related to MFAR by:

$$q = \mathbf{1} - \text{MFAR}.\tag{34}$$

Thus, besides introducing more iterations and computation load, higher MFAR will also lead to a smaller h-subset, which indicates more localization and less

Qx≔Minimize ∑

8 >>><

>>>:

(rx 2

8 >< >:

34

Qy≔Minimize ∑

2

2

extended Fast-LTS (EF-LTS): Algorithm 3: EF-LTS

> r0x <sup>¼</sup> ½ � <sup>r</sup>0x1;r0x2; <sup>⋯</sup>;r0xn <sup>T</sup> r0y ¼ r0y1;r0y2; ⋯;r0yn

Hx0 ¼ f g πxð Þ1 ; πxð Þ2 ; ⋯; πxð Þ h ⊂f g 1; 2; ⋯; n Hy0 <sup>¼</sup> <sup>π</sup>yð Þ<sup>1</sup> ; <sup>π</sup>yð Þ<sup>2</sup> ; <sup>⋯</sup>; <sup>π</sup>yð Þ <sup>h</sup> � �⊂f g <sup>1</sup>; <sup>2</sup>; <sup>⋯</sup>; <sup>n</sup> <sup>s</sup>:t:

obtain a set of h-subsets Hx2 and Hy2.

rrx <sup>¼</sup> ½ � rrx1;rrx2; <sup>⋯</sup>;rrxn <sup>T</sup> rry ¼ rry1;rry2; ⋯;rryn

( (

and estimate the error scales σ<sup>x</sup> and σ<sup>y</sup> by

� �<sup>T</sup> with

( (

� �<sup>T</sup> and

Then construct the initial h-subsets Hx0 and Hy0 by:

)<sup>i</sup> ≤ ��� ≤ (rx

where (rx

)1 ≤ ��� ≤ (rx

h i¼1 r2 x � � i

h i¼1 r2 y � � i with rx <sup>¼</sup> ½ � rx1;rx2; <sup>⋯</sup>;rxn <sup>T</sup> ry ¼ ry1;ry2; ⋯;ryn

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

8 < :

2

� �<sup>T</sup> and

)<sup>i</sup> represents the ith element of the ordered squared residuals

)n, and the meaning of (ry

Step 31. Randomly draw p feature matches and LS fit them to estimate the initial

<sup>r</sup>0xi <sup>¼</sup> xsi � <sup>X</sup><sup>T</sup>

<sup>r</sup>0yi <sup>¼</sup> ysi � <sup>X</sup><sup>T</sup>

r2 0x � �

8 >< >:

Carry out two C-steps on Hx0 and Hy0 to obtain the h-subsets Hx2 and Hy2 with

Step 32. Select 10 Hx2 with the smallest 10 Qx and 10 Hy2 with the smallest 10 Qy if T is larger than 10; otherwise, select all Hx2 and Hy2. Carry out C-steps on these h-subsets until convergence. The solutions corresponding to the smallest Qx

smaller Qx and Qy, respectively. Iteratively repeat above procedures T times to

and Qy are selected as the raw estimations θ<sup>r</sup> and ψr, respectively. Step 33. Calculate residuals rrx and rry based on θ<sup>r</sup> and ψr,

σ<sup>x</sup> ¼ C<sup>1</sup>

8 >>>>><

>>>>>:

σ<sup>y</sup> ¼ C<sup>2</sup>

distributions [50]. Based on (27), we further calculate two weights by:

s

s

where C<sup>1</sup> and C<sup>2</sup> are correction factors to achieve consistency at Gaussian error

r2 0y � �

rrxi <sup>¼</sup> xsi � <sup>X</sup><sup>T</sup>

rryi <sup>¼</sup> ysi � <sup>X</sup><sup>T</sup>

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 h ∑ h i¼1 r2 rx � � i

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 h ∑ h i¼1 r2 ry � � i <sup>i</sup> θ<sup>r</sup>

<sup>i</sup> ψ<sup>r</sup>

<sup>i</sup> θ<sup>0</sup>

, i ¼ 1, …, n:

<sup>π</sup>xð Þ <sup>h</sup> <sup>≤</sup> … <sup>≤</sup> r2

<sup>π</sup>yð Þ <sup>h</sup> <sup>≤</sup> … <sup>≤</sup> <sup>r</sup><sup>2</sup>

0x � � πxð Þ n

> 0y � �

πyð Þ n :

(25)

(26)

(27)

0x � �

> 0y � �

, i ¼ 1, …, n

<sup>i</sup> ψ<sup>0</sup>

<sup>π</sup>xð Þ<sup>1</sup> ≤⋯≤ <sup>r</sup><sup>2</sup>

<sup>π</sup>yð Þ<sup>1</sup> ≤⋯≤ <sup>r</sup><sup>2</sup>

parameters θ<sup>0</sup> and ψ0, and calculate the initial residuals r0x and r0y by

Each of the two optimizations in (23) is of the standard form (20). A direct solution to (23) may be thus achieved by decomposing 2D regression as two independent 1D regressions and using Fast-LTS to conduct estimation, respectively. This idea is feasible, but it may result in unnecessary computations because the feature positions in two image directions are in fact tied to each other, i.e., for the ith feature (xi, yi), the selection of xi will naturally mean the selection of yi. We can thus combine the two 1D regressions into a real 2D regression effectively, i.e., the

rxi <sup>¼</sup> xsi � <sup>X</sup><sup>T</sup>

8 < :

2

ryi <sup>¼</sup> ysi � <sup>X</sup><sup>T</sup>

i θ

, i ¼ 1, …, n

(23)

(24)

<sup>i</sup> ψ

)<sup>i</sup> can be likewise inferred.


Algorithm 4: Accurate SAR image registration based on SURF features and

Step 41. Use FH-9(-Fs) to extract SURF keypoints from master and slave

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

Step 42. Construct initial feature correspondences by simply matching SURF

Step 43. Robustly processing the correspondences with EF-LTS to retrieve the

Step 44. Transform and interpolate the slave image to geometrically align it to

Actually, this scheme has been put into practice in the above experiments. In this section, we further devise an experiment to check it on MiniSAR pair. The images we use are two high-resolution SAR images of the entrance gate of the Sandia Research Park acquired by the Ku-Band MiniSAR system developed by the Sandia Laboratory [51]. The images are taken from different tracks with different incidences and squints, as listed in Table 5, while the platform altitude is just beyond 1 km. All these reveal the nontrivial target relief-induced geometrical warping between images, which, however, cannot be compensated beforehand for lack of ground truth such as DEM and target height. Besides this, the images also experience a very large intensity variation. To enhance the texture, we use the logarithmic intensity of original complex images, as shown in Figure 7(a) and (b). To achieve a more precise approximation to the real warping, we divide the image pair into four 500 500 patch pairs. The geometrical warping on each patch pair is approximated as an affine transformation (the higher order polynomial has also been used to model the warp function, but unsatisfactory registration result is attained). We adopt HF-9(-4) SURF detector to extract feature correspondences from each patch pair, and EF-LTS is then used to obtain the affine parameters, based on which the slave image is finally aligned to the master image. To illustrate the registration accuracy, we fuse and overlap the coregistered images together. The

RGB fusion in Figure 7(c) is obtained by treating the master image and the coregistered slave image as red and green, respectively, while zeroing the blue component. The well-distributed yellow then immediately illustrates the accurate registration of the images. The overlapping in Figure 7(d) is obtained by simply averaging the two coregistered images. It contains the whole information of the two

we focus on the two pole-like target areas 1 and 2 in Figure 7(d) with their corresponding Google optical images shown in Figure 8(g) and (h), respectively. Figure 8(i) portrays the details of Pole 2 in the Street View of Google Maps. The

To further evaluate the registration performance of the scheme, in the following

Parameters Master image Slave image Azimuth resolution 0.1016 m 0.1016 m Range resolution 0.1016 m 0.1016 m Grazing angle 27.0107° 26.1892° Global track angle 158.3687° 153.0825° Central frequency 16.8 GHz 16.8 GHz Platform altitude 1.6715 km 1.6715 km Squint 89.9935° 89.9924°

EF-LTS

descriptors.

warp function.

master image.

images, respectively.

DOI: http://dx.doi.org/10.5772/intechopen.81665

images but has fewer speckles.

Imaging parameters of the two MiniSAR images.

Table 5.

37

Table 4.

Sampling number T under different inlier percentage q and polynomial order N when ε = 0.99.

accuracy in estimation and worse consistency between the extracted features and retrieval parameters. This is why FH-15(-2) can achieve better ATE than FH-9(-1), as displayed in Table 2. As presented in Section 3, on MFAR and many other criteria, SURF is identified to be the best for general SAR image registration. SURF may thus also improve the efficiency and accuracy of parameter retrieval besides the good performance on feature extraction and matching.

When the correspondence number n is large, a similar nested extension can be also taken for EF-LTS by randomly partitioning the correspondences into M subsets with equal cardinality, and the trimming constant hs and sampling number Ts of each subset should be also reduced by M times relative to h and T. On each subset, we first implement Step 31 for Ts hs-subsets of Hx2 and Hy2. Based on which we then implement Step 32 and Step 33 on all the constructed correspondences with original h and T. In this way, an efficient retrieval can be still achieved.

To evaluate EF-LTS for SAR image registration, we also use it to the InSAR image pair given in Figure 4(a) and (b). Similarly, the feature correspondences are first constructed by SURF with HF-9(-1), then we run EF-LTS 100 times to retrieve the affine parameters and calculate CC and SNR. The obtained parameters, CC, and SNR of each execution are shown in Figure 5, while the mean and standard deviation of the parameters, CC, and SNR are listed in Table 3. It is revealed that EF-LTS behaves very stable and the estimated parameters, CC, and SNR are invariant for each execution. It can reach an averagely better CC and SNR than RANSAC and is more appropriate for InSAR image registration. Figure 4(c) and (d) further illustrates the interferogram and correlation map of the coregistered InSAR pair with wrap parameters estimated by EF-LTS. Interferogram is the argument or phase of the dot production between the complex master image and the complex conjugation of the registered slave image, while correlation map measures CC of the 3 3 patches around each corresponding pixel position between the images. The interferogram fringe is clear and the correlation is strong in stable area such as the brighter buildings in Figure 4(a) and (b) and the upper-right bare land. But in the upper-left residential area, the interferogram becomes less clear and the correlation is relatively small probably because the scattering is very sensitive to incidence changes. While in other area (mainly vegetable lands and parking lot), the interferogram is almost lost and the coherence is very low due to the temporal and/or volume decorrelation. All these match with the ground truth very well.

#### 5. Experiment and analysis

Based on the finding in Sections 2–4, we propose to conduct high accurate SAR image registration by using EF-LTS to fit the SURF correspondences. The scheme works as follows:

## Algorithm 4: Accurate SAR image registration based on SURF features and EF-LTS

Step 41. Use FH-9(-Fs) to extract SURF keypoints from master and slave images, respectively.

Step 42. Construct initial feature correspondences by simply matching SURF descriptors.

Step 43. Robustly processing the correspondences with EF-LTS to retrieve the warp function.

Step 44. Transform and interpolate the slave image to geometrically align it to master image.

Actually, this scheme has been put into practice in the above experiments. In this section, we further devise an experiment to check it on MiniSAR pair. The images we use are two high-resolution SAR images of the entrance gate of the Sandia Research Park acquired by the Ku-Band MiniSAR system developed by the Sandia Laboratory [51]. The images are taken from different tracks with different incidences and squints, as listed in Table 5, while the platform altitude is just beyond 1 km. All these reveal the nontrivial target relief-induced geometrical warping between images, which, however, cannot be compensated beforehand for lack of ground truth such as DEM and target height. Besides this, the images also experience a very large intensity variation. To enhance the texture, we use the logarithmic intensity of original complex images, as shown in Figure 7(a) and (b). To achieve a more precise approximation to the real warping, we divide the image pair into four 500 500 patch pairs. The geometrical warping on each patch pair is approximated as an affine transformation (the higher order polynomial has also been used to model the warp function, but unsatisfactory registration result is attained). We adopt HF-9(-4) SURF detector to extract feature correspondences from each patch pair, and EF-LTS is then used to obtain the affine parameters, based on which the slave image is finally aligned to the master image. To illustrate the registration accuracy, we fuse and overlap the coregistered images together. The RGB fusion in Figure 7(c) is obtained by treating the master image and the coregistered slave image as red and green, respectively, while zeroing the blue component. The well-distributed yellow then immediately illustrates the accurate registration of the images. The overlapping in Figure 7(d) is obtained by simply averaging the two coregistered images. It contains the whole information of the two images but has fewer speckles.

To further evaluate the registration performance of the scheme, in the following we focus on the two pole-like target areas 1 and 2 in Figure 7(d) with their corresponding Google optical images shown in Figure 8(g) and (h), respectively. Figure 8(i) portrays the details of Pole 2 in the Street View of Google Maps. The


Table 5.

Imaging parameters of the two MiniSAR images.

accuracy in estimation and worse consistency between the extracted features and retrieval parameters. This is why FH-15(-2) can achieve better ATE than FH-9(-1), as displayed in Table 2. As presented in Section 3, on MFAR and many other criteria, SURF is identified to be the best for general SAR image registration. SURF may thus also improve the efficiency and accuracy of parameter retrieval besides

Sampling number T under different inlier percentage q and polynomial order N when ε = 0.99.

0 7 64 4 32 2 35 19 11 9 7 4 3 293 97 37 24 16 7 4 4714 760 161 80 41 11 6

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

0.5 0.6 0.7 0.75 0.8 0.9 0.95

When the correspondence number n is large, a similar nested extension can be also taken for EF-LTS by randomly partitioning the correspondences into M subsets with equal cardinality, and the trimming constant hs and sampling number Ts of each subset should be also reduced by M times relative to h and T. On each subset, we first implement Step 31 for Ts hs-subsets of Hx2 and Hy2. Based on which we then implement Step 32 and Step 33 on all the constructed correspondences with

To evaluate EF-LTS for SAR image registration, we also use it to the InSAR image pair given in Figure 4(a) and (b). Similarly, the feature correspondences are first constructed by SURF with HF-9(-1), then we run EF-LTS 100 times to retrieve the affine parameters and calculate CC and SNR. The obtained parameters, CC, and SNR of each execution are shown in Figure 5, while the mean and standard deviation of the parameters, CC, and SNR are listed in Table 3. It is revealed that EF-LTS behaves very stable and the estimated parameters, CC, and SNR are invariant for each execution. It can reach an averagely better CC and SNR than RANSAC and is more appropriate for InSAR image registration. Figure 4(c) and (d) further illustrates the interferogram and correlation map of the coregistered InSAR pair with wrap parameters estimated by EF-LTS. Interferogram is the argument or phase of the dot production between the complex master image and the complex conjugation of the registered slave image, while correlation map measures CC of the 3 3 patches around each corresponding pixel position between the images. The interferogram fringe is clear and the correlation is strong in stable area such as the brighter buildings in Figure 4(a) and (b) and the upper-right bare land. But in the upper-left residential area, the interferogram becomes less clear and the correlation is relatively small probably because the scattering is very sensitive to incidence changes. While in other area (mainly vegetable lands and parking lot), the interferogram is almost lost and the coherence is very low due to the temporal and/or volume decorrelation. All these match with the ground truth very well.

Based on the finding in Sections 2–4, we propose to conduct high accurate SAR image registration by using EF-LTS to fit the SURF correspondences. The scheme

the good performance on feature extraction and matching.

N q

Table 4.

5. Experiment and analysis

works as follows:

36

original h and T. In this way, an efficient retrieval can be still achieved.

#### Figure 7.

Registration of the MiniSAR image pair. (a) Master image, (b) slave image, and (c) pseudocolor fusion as well as (d) the overlapping of them after registration with EF-LTS to fit the SURF features. "1" and "2" in (d) indicate two pole-like targets which are further detailed in Figure 8.

monopolarized, the developed scheme is also appropriate to the registration of fully polarimetric SAR (PolSAR) images. Different from monopolarized SAR, each cell in PolSAR image is a scattering matrix S with four entries SHH, SHV, SVH, and

Registration of the two pole-like targets. SAR imagery of pole "1" in (a) master image, (b) coregistered slave image, and (c) overlapped image, as well as (g) the corresponding Google optical image. SAR imagery of pole "2" in (d) master image, (e) coregistered slave image, and (f) overlapped image, as well as (h) its Google

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

<sup>S</sup> <sup>¼</sup> SHH SHV

<sup>2</sup> <sup>þ</sup> j j SHV

we can then obtain the total power (also known as SPAN) of target. An accurate

<sup>2</sup> <sup>þ</sup> j j SVH

Nevertheless, by taking the squared Frobenius norm of matrix S [53]:

registration of PolSAR images can be eventually achieved by simply using the

SAR coherent imaging unavoidably brings about geometrical distortion and speckle into the acquired images and makes the registration of SAR images much more complicated. In this chapter, we focus on two important procedures in general feature-based SAR registration, i.e., the feature extraction and the parameter retrieval by identifying the appropriate feature and the appropriate estimation algorithm. As for the former, we conduct a detailed evaluation on the commonly used features such as tie points, Harris corner, SIFT, and SURF. We find that SURF outperforms others in terms of the geometrical invariance of feature, extraction speed, accuracy of localization, geometrical invariance of descriptor, matching speed, robustness to decorrelation, and flexibility to image speckling. Among these criteria, feature's flexibility to speckle is particularly focused because speckle impacts the feature extraction and matching, while speckle filtering may change the feature position and impact the subpixel localization. The Fast-Hessian detector of

<sup>F</sup> ¼ j j SHH

SPAN <sup>¼</sup> k k<sup>S</sup> <sup>2</sup>

optical image and (i) detailed portrayal in Google Maps.

DOI: http://dx.doi.org/10.5772/intechopen.81665

developed scheme to the corresponding SPAN image pair.

SVH SVV : (35)

<sup>2</sup> <sup>þ</sup> j j SVV

<sup>2</sup> (36)

SVV [52]:

Figure 8.

6. Conclusion

39

target is shown to be the power transmission pole. Figure 8(a)–(c) exhibits the SAR imagery of Pole 1 in the master image, coregistered slave image, and overlapped image, respectively. The corresponding SAR imageries of Pole 2 are displayed in Figure 8(d)–(f), respectively. It is known that the darker pole-like feature in each SAR image is not the real pole scattering, but its shadow under the irradiation of radar. The actual scattering center of the pole is overlapped with its ground position because of the dominant dihedral backscattering between the pole and ground. From Figure 8(c) and (f), we can find that the shadows of the two poles are still separated after registration due to the volume-induced warping. According to our estimate, the separations are about 6.5 and 5°, respectively, which approach to the actual track angle 5.2862°. Except for these shadows, the poles and other area are accurately overlapped. Nice registration is still achieved despite the large local distortion and decorrelation. Moreover, the experiment also validates the strategy for general feature-based SAR image registration, i.e., to focus on the global registration and to neglect the local discontentment. The accurate registration of each pixel is impossible and unnecessary. It should be noted that the conventional SAR image registrations including the feature-based approaches focused in current chapter are mainly appropriate for images with approximated low-order polynomial geometrical warping. For SAR images taken from area of rough topography with long baseline, we need some more complex approaches with the a priori ground truth information being included, such as the DEM-assisted registration [48]. Although the SAR and InSAR image pairs used in the experiment are all

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

#### Figure 8.

Registration of the two pole-like targets. SAR imagery of pole "1" in (a) master image, (b) coregistered slave image, and (c) overlapped image, as well as (g) the corresponding Google optical image. SAR imagery of pole "2" in (d) master image, (e) coregistered slave image, and (f) overlapped image, as well as (h) its Google optical image and (i) detailed portrayal in Google Maps.

monopolarized, the developed scheme is also appropriate to the registration of fully polarimetric SAR (PolSAR) images. Different from monopolarized SAR, each cell in PolSAR image is a scattering matrix S with four entries SHH, SHV, SVH, and SVV [52]:

$$\mathbf{S} = \begin{bmatrix} \mathbf{S}\_{HH} & \mathbf{S}\_{HV} \\ \mathbf{S}\_{VH} & \mathbf{S}\_{VV} \end{bmatrix}. \tag{35}$$

Nevertheless, by taking the squared Frobenius norm of matrix S [53]:

$$\text{SPAN} = \left\| \mathbf{S} \right\|\_{F}^{2} = \left| \mathbf{S}\_{HH} \right|^{2} + \left| \mathbf{S}\_{HV} \right|^{2} + \left| \mathbf{S}\_{VH} \right|^{2} + \left| \mathbf{S}\_{VV} \right|^{2} \tag{36}$$

we can then obtain the total power (also known as SPAN) of target. An accurate registration of PolSAR images can be eventually achieved by simply using the developed scheme to the corresponding SPAN image pair.

#### 6. Conclusion

target is shown to be the power transmission pole. Figure 8(a)–(c) exhibits the SAR imagery of Pole 1 in the master image, coregistered slave image, and overlapped image, respectively. The corresponding SAR imageries of Pole 2 are displayed in Figure 8(d)–(f), respectively. It is known that the darker pole-like feature in each SAR image is not the real pole scattering, but its shadow under the irradiation of radar. The actual scattering center of the pole is overlapped with its ground position because of the dominant dihedral backscattering between the pole and ground. From Figure 8(c) and (f), we can find that the shadows of the two poles are still separated after registration due to the volume-induced warping. According to our estimate, the separations are about 6.5 and 5°, respectively, which approach to the actual track angle 5.2862°. Except for these shadows, the poles and other area are accurately overlapped. Nice registration is still achieved despite the large local distortion and decorrelation. Moreover, the experiment also validates the strategy for general feature-based SAR image registration, i.e., to focus on the global registration and to neglect the local discontentment. The accurate registration of each pixel is impossible and unnecessary. It should be noted that the conventional SAR image registrations including the feature-based approaches focused in current chapter are mainly appropriate for images with approximated low-order polynomial geometrical warping. For SAR images taken from area of rough topography with long baseline, we need some more complex approaches with the a priori ground truth information being included, such as the DEM-assisted registration [48]. Although the SAR and InSAR image pairs used in the experiment are all

Registration of the MiniSAR image pair. (a) Master image, (b) slave image, and (c) pseudocolor fusion as well as (d) the overlapping of them after registration with EF-LTS to fit the SURF features. "1" and "2" in (d)

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

indicate two pole-like targets which are further detailed in Figure 8.

Figure 7.

38

SAR coherent imaging unavoidably brings about geometrical distortion and speckle into the acquired images and makes the registration of SAR images much more complicated. In this chapter, we focus on two important procedures in general feature-based SAR registration, i.e., the feature extraction and the parameter retrieval by identifying the appropriate feature and the appropriate estimation algorithm. As for the former, we conduct a detailed evaluation on the commonly used features such as tie points, Harris corner, SIFT, and SURF. We find that SURF outperforms others in terms of the geometrical invariance of feature, extraction speed, accuracy of localization, geometrical invariance of descriptor, matching speed, robustness to decorrelation, and flexibility to image speckling. Among these criteria, feature's flexibility to speckle is particularly focused because speckle impacts the feature extraction and matching, while speckle filtering may change the feature position and impact the subpixel localization. The Fast-Hessian detector of

SURF has a potential relation with the refined Lee speckle filter. SSP in SURF just indicates that we use a series of box filters of different size to filter speckles and extract features of different scales. Thus, SURF is very flexible to deal with SAR speckle. In view of the application with strict requirement for registration accuracy, we suggest using the SURF detector of HF-9(-1) to the Fs times interpolated images with unchanged sampling step to extract feature. The new detector HF-9(-Fs) can significantly improve the registration accuracy to subpixel (<1 pixel) and is especially fit for high accurate SAR image registration.

Parameter retrieval in SAR registration is difficult because spatial or temporal decorrelation will always introduce mismatches into the obtained feature correspondences. The estimator should be robust to outliers. We find that the commonly used RANSAC may trap into local occlusion and result in uncertain parameter retrieval. This uncertainty is more severe in SAR image registration because SAR geometrical warping varies from pixel to pixel, but the low-order polynomial approximation can only account for global registration instead of the local contentment. The local-to-global strategy in RANSAC may thus magnify the local distortion, aggravate the estimation uncertainty, and damnify the global registration accuracy although a largest CS is obtained. To achieve a stable registration for SAR images, we should estimate the parameters with more correspondences to reflect the true support than just a MSS, and apply an appropriate loss function. This leads us to EF-LTS, which improves Fast-LTS from 1D regression to 2D regression, and provides us an adaptive determination of the number of random sampling instead of setting it as a constant 500. EF-LTS conducts registration by LS fitting at least half of the correspondences to minimize the squared residual. It behaves very stable and is averagely better than RANSAC. Hence, we recommend conducting SAR image registration by fitting SURF features with EF-LTS. Experiments on both InSAR and MiniSAR image pairs validate the nice performance of this registration scheme.

## Acknowledgements

This work is supported by China Manned Space Program along with the Youth Innovation Promotion Association, Chinese Academy of Sciences under Grant No. 2014131. The authors thank the International Society for Optics and Photonics (SPIE) for the permission to reuse materials that have appeared in Proceedings of SPIE (Li D, Zhang Y. On the appropriate feature for general SAR image registration; The appropriate parameter retrieval algorithm for feature-based SAR image registration. SAR Image Analysis Modeling and Techniques XII. Vol. 8536, 2012.)

Author details

, Yunhua Zhang1,2\* and Xiaojin Shi1

2 University of Chinese Academy of Sciences, Beijing, China

\*Address all correspondence to: zhangyunhua@mirslab.cn

Chinese Academy of Sciences, Beijing, China

provided the original work is properly cited.

1 Key Laboratory of Microwave Remote Sensing, National Space Science Center,

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

DOI: http://dx.doi.org/10.5772/intechopen.81665

© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium,

Dong Li<sup>1</sup>

41

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

## Author details

SURF has a potential relation with the refined Lee speckle filter. SSP in SURF just indicates that we use a series of box filters of different size to filter speckles and extract features of different scales. Thus, SURF is very flexible to deal with SAR speckle. In view of the application with strict requirement for registration accuracy, we suggest using the SURF detector of HF-9(-1) to the Fs times interpolated images with unchanged sampling step to extract feature. The new detector HF-9(-Fs) can significantly improve the registration accuracy to subpixel (<1 pixel) and is espe-

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Parameter retrieval in SAR registration is difficult because spatial or temporal decorrelation will always introduce mismatches into the obtained feature correspondences. The estimator should be robust to outliers. We find that the commonly used RANSAC may trap into local occlusion and result in uncertain parameter retrieval. This uncertainty is more severe in SAR image registration because SAR geometrical warping varies from pixel to pixel, but the low-order polynomial approximation can only account for global registration instead of the local contentment. The local-to-global strategy in RANSAC may thus magnify the local distortion, aggravate the estimation uncertainty, and damnify the global registration accuracy although a largest CS is obtained. To achieve a stable registration for SAR images, we should estimate the parameters with more correspondences to reflect the true support than just a MSS, and apply an appropriate loss function. This leads us to EF-LTS, which improves Fast-LTS from 1D regression to 2D regression, and provides us an adaptive determination of the number of random sampling instead of setting it as a constant 500. EF-LTS conducts registration by LS fitting at least half of the correspondences to minimize the squared residual. It behaves very stable and is averagely better than RANSAC. Hence, we recommend conducting SAR image registration by fitting SURF features with EF-LTS. Experiments on both InSAR and MiniSAR image pairs validate the nice performance of this registration

This work is supported by China Manned Space Program along with the Youth Innovation Promotion Association, Chinese Academy of Sciences under Grant No. 2014131. The authors thank the International Society for Optics and Photonics (SPIE) for the permission to reuse materials that have appeared in Proceedings of SPIE (Li D, Zhang Y. On the appropriate feature for general SAR image registration; The appropriate parameter retrieval algorithm for feature-based SAR image registration. SAR Image Analysis Modeling and Techniques XII. Vol. 8536, 2012.)

cially fit for high accurate SAR image registration.

scheme.

40

Acknowledgements

Dong Li<sup>1</sup> , Yunhua Zhang1,2\* and Xiaojin Shi1

1 Key Laboratory of Microwave Remote Sensing, National Space Science Center, Chinese Academy of Sciences, Beijing, China

2 University of Chinese Academy of Sciences, Beijing, China

\*Address all correspondence to: zhangyunhua@mirslab.cn

© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

## References

[1] Homer J, Longstaff ID. Minimising the tie patch window size for SAR image coregistration. Electronics Letters. 2003; 39:122-124. DOI: 10.1049/el:20030054

[2] Zou W, Li Y, Li Z, Ding X. Improvement of the accuracy of InSAR image co-registration based on tie points: A review. Sensors. 2009;9: 1259-1281. DOI: 10.3390/s90201259

[3] Giles AB, Massom RA, Warner RC. A method for sub-pixel scale featuretracking using Radarsat images applied to the Mertz Glacier Tongue, East Antarctica. Remote Sensing of Environment. 2009;113:1691-1699. DOI: 10.1016/j.rse.2009.03.015

[4] Li D, Zhang Y. A fast offset estimation approach for InSAR image subpixel registration. IEEE Geoscience and Remote Sensing Letters. 2012;9: 267-271. DOI: 10.1109/LGRS.2011. 2166752

[5] Harris C, Stephens M. A combined corner and edge detector. In: Proceedings of the Alvey Vision Conference (AVC'88); 31 August–2 September 1998; Manchester, U.K. pp. 147-151

[6] Bentoutou Y, Taleb N, Kpalma K, Ronsin J. An automatic image registration for applications in remote sensing. IEEE Transactions on Geoscience and Remote Sensing. 2005; 43:2127-2137. DOI: 10.1109/TGRS. 2005.853187

[7] Lowe DG. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision. 2004;60:91-110. DOI: 10.1023/B: VISI.0000029664.99615.94

[8] Bay H, Ess A, Tuytelaars T, Van Gool L. Speeded-up robust features (SURF).

Computer Vision and Image Understanding. 2008;110:346-359. DOI: 10.1016/j.cviu.2007.09.014

Remote Sensing. 2010;31:1959-1980. DOI: 10.1080/01431160902927622

DOI: http://dx.doi.org/10.5772/intechopen.81665

Geoscience and Remote Sensing Letters. 2017;14:3-7. DOI: 10.1109/LGRS.

[23] Xiang Y, Wang F, You H. OS-SIFT: A robust SIFT-like algorithm for highresolution optical-to-SAR image registration in suburban areas. IEEE Transactions on Geoscience and Remote Sensing. 2018;56:3078-3090. DOI: 10.1109/TGRS.2018.2790483

[24] Teke M, Temizel A. Multi-spectral satellite image registration using scalerestricted SURF. In: Proceedings of International Conference on Pattern Recognition (ICPR'10); 23–26 August 2010; Istanbul, Turkey. pp. 2310-2313

[25] Nguyen H-G, Fablet R, Ehrhold A, Boucher J-M. Keypoint-based analysis of sonar images: Applications to seabed recognition. IEEE Transactions on Geoscience and Remote Sensing. 2012;

50:1171-1184. DOI: 10.1109/

[26] Liu R, Wang Y. SAR image matching based on speeded up robust feature. In: Proceedings of the WRI Global Congress on Intelligent System (GCIS'09); 19–21 May, 2009; Xiamen,

[27] Lee WK, Kim AL. An efficient automatic geo-registration technique for high resolution spaceborne SAR image

fusion. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGASS'11); 24–29

July, 2011; Vancouver, Canada.

[28] Zhang H, Ni W, Yan W, Wu J, Li S. Robust SAR image registration based on edge matching and refined coherent point drift. IEEE Geoscience and Remote Sensing Letters. 2015;12: 2115-2119. DOI: 10.1109/LGRS.

TGRS.2011.2165848

China. pp. 518-522

pp. 3566-3569

2015.2451396

2016.2600858

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

[16] Wang S, You H, Fu K. BFSIFT: A novel method to find feature matches for SAR image registration. IEEE Geoscience and Remote Sensing Letters.

2012;9:649-653. DOI: 10.1109 /

[17] Gong M, Zhao S, Jiao L, Tian D, Wang S. A novel coarse-to-fine scheme for automatic image registration based on SIFT and mutual Information. IEEE Transactions on Geoscience and Remote Sensing. 2014;52:4328-4338. DOI: 10.1109/TGRS.2013.2281391

[18] Dellinger F, Delon J, Gousseau Y, Michel J, Tupin F. SAR-SIFT: A SIFTlike algorithm for SAR images. IEEE Transactions on Geoscience and Remote Sensing. 2015;53:453-466. DOI: 10.1109/

[19] Wang F, You H, Fu X. Adapted anisotropic Gaussian SIFT matching strategy for SAR registration. IEEE Geoscience and Remote Sensing Letters. 2015;12:160-164. DOI: 10. 1109/LGRS.

[20] Wang B, Zhang J, Lu L, Huang G, Zhao Z. A uniform SIFT-like algorithm for SAR image registration. IEEE Geoscience and Remote Sensing Letters. 2015;12:1426-1430. DOI: 10.1109/

[21] Zeng L, Zhou D, Liang J, Zhang K. Polar scale-invariant feature transform for synthetic aperture radar image registration. IEEE Geoscience and Remote Sensing Letters. 2017;14: 1101-1105. DOI: 10.1109/LGRS.

[22] Ma W, Wen Z, Wu Y, Jiao L, Gong M, Zheng Y, et al. Remote sensing image registration with modified SIFT and enhanced feature matching. IEEE

LGRS.2011.2177437

TGRS.2014.2323552

2014.2330593

LGRS.2015.2406336

2017.2698450

43

[9] Lindeberg T. Feature detection with automatic scale selection. Internatioal Journal of Computer Vision. 1998;30: 79-116. DOI: 10.1023/A:1008045108935

[10] Mikolajczyk K, Schmid C. A performance evaluation of local descriptors. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2005;27:1615-1630. DOI: 10. 1109/TPAMI.2005.188

[11] Wessel B, Huber M, Roth A. Automatic image-to-image registration of near real-time SAR images. In: Proceedings of the Envisat Symposium; 23–27 April 2007; Montreux, Switzerland. pp. 1-6

[12] Li F, Zhang G, Yan J. Coregistration based on SIFT algorithm for synthetic aperture radar interferometry. In: Archives of International Society got Photogrammetry and Remote Sensing (ISPRS'08); 3–11 July 2008; Beijing, China. pp. 123-128

[13] Chen E, Li Z, Tian X, Li S. Application of scale invariant feature transformation to SAR imagery registration. Acta Automatica Sinica. 2008;34:861-868. DOI: 10.3724/SP. J.1004. 2008.00861

[14] Li D, Zhang Y. Geometric featurebased image coregistration approach for InSAR. In: Proceedings of the Asian-Pacific Conference on Synthetic Aperture Radar; 26–30 October, 2009; Xi'an, China. pp. 1026-1030

[15] Schwind P, Suri S, Reinartz P, Siebert A. Applicability of the SIFT operator to geometric SAR image registration. International Journal of On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

Remote Sensing. 2010;31:1959-1980. DOI: 10.1080/01431160902927622

References

[1] Homer J, Longstaff ID. Minimising the tie patch window size for SAR image coregistration. Electronics Letters. 2003; 39:122-124. DOI: 10.1049/el:20030054

Computer Vision and Image

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

10.1016/j.cviu.2007.09.014

Understanding. 2008;110:346-359. DOI:

[9] Lindeberg T. Feature detection with automatic scale selection. Internatioal Journal of Computer Vision. 1998;30: 79-116. DOI: 10.1023/A:1008045108935

[10] Mikolajczyk K, Schmid C. A performance evaluation of local descriptors. IEEE Transactions on Pattern Analysis and Machine

[11] Wessel B, Huber M, Roth A. Automatic image-to-image registration of near real-time SAR images. In: Proceedings of the Envisat Symposium;

23–27 April 2007; Montreux,

[13] Chen E, Li Z, Tian X, Li S. Application of scale invariant feature transformation to SAR imagery registration. Acta Automatica Sinica. 2008;34:861-868. DOI: 10.3724/SP.

Xi'an, China. pp. 1026-1030

[15] Schwind P, Suri S, Reinartz P, Siebert A. Applicability of the SIFT operator to geometric SAR image registration. International Journal of

[14] Li D, Zhang Y. Geometric featurebased image coregistration approach for InSAR. In: Proceedings of the Asian-Pacific Conference on Synthetic Aperture Radar; 26–30 October, 2009;

Switzerland. pp. 1-6

China. pp. 123-128

J.1004. 2008.00861

10. 1109/TPAMI.2005.188

Intelligence. 2005;27:1615-1630. DOI:

[12] Li F, Zhang G, Yan J. Coregistration based on SIFT algorithm for synthetic aperture radar interferometry. In: Archives of International Society got Photogrammetry and Remote Sensing (ISPRS'08); 3–11 July 2008; Beijing,

Improvement of the accuracy of InSAR image co-registration based on tie points: A review. Sensors. 2009;9: 1259-1281. DOI: 10.3390/s90201259

[3] Giles AB, Massom RA, Warner RC. A method for sub-pixel scale featuretracking using Radarsat images applied to the Mertz Glacier Tongue, East Antarctica. Remote Sensing of

Environment. 2009;113:1691-1699. DOI:

[5] Harris C, Stephens M. A combined

[6] Bentoutou Y, Taleb N, Kpalma K, Ronsin J. An automatic image

registration for applications in remote

Geoscience and Remote Sensing. 2005; 43:2127-2137. DOI: 10.1109/TGRS.

[7] Lowe DG. Distinctive image features

[8] Bay H, Ess A, Tuytelaars T, Van Gool L. Speeded-up robust features (SURF).

sensing. IEEE Transactions on

from scale-invariant keypoints. International Journal of Computer Vision. 2004;60:91-110. DOI: 10.1023/B:

VISI.0000029664.99615.94

corner and edge detector. In: Proceedings of the Alvey Vision Conference (AVC'88); 31 August–2 September 1998; Manchester, U.K.

10.1016/j.rse.2009.03.015

2166752

pp. 147-151

2005.853187

42

[4] Li D, Zhang Y. A fast offset estimation approach for InSAR image subpixel registration. IEEE Geoscience and Remote Sensing Letters. 2012;9: 267-271. DOI: 10.1109/LGRS.2011.

[2] Zou W, Li Y, Li Z, Ding X.

[16] Wang S, You H, Fu K. BFSIFT: A novel method to find feature matches for SAR image registration. IEEE Geoscience and Remote Sensing Letters. 2012;9:649-653. DOI: 10.1109 / LGRS.2011.2177437

[17] Gong M, Zhao S, Jiao L, Tian D, Wang S. A novel coarse-to-fine scheme for automatic image registration based on SIFT and mutual Information. IEEE Transactions on Geoscience and Remote Sensing. 2014;52:4328-4338. DOI: 10.1109/TGRS.2013.2281391

[18] Dellinger F, Delon J, Gousseau Y, Michel J, Tupin F. SAR-SIFT: A SIFTlike algorithm for SAR images. IEEE Transactions on Geoscience and Remote Sensing. 2015;53:453-466. DOI: 10.1109/ TGRS.2014.2323552

[19] Wang F, You H, Fu X. Adapted anisotropic Gaussian SIFT matching strategy for SAR registration. IEEE Geoscience and Remote Sensing Letters. 2015;12:160-164. DOI: 10. 1109/LGRS. 2014.2330593

[20] Wang B, Zhang J, Lu L, Huang G, Zhao Z. A uniform SIFT-like algorithm for SAR image registration. IEEE Geoscience and Remote Sensing Letters. 2015;12:1426-1430. DOI: 10.1109/ LGRS.2015.2406336

[21] Zeng L, Zhou D, Liang J, Zhang K. Polar scale-invariant feature transform for synthetic aperture radar image registration. IEEE Geoscience and Remote Sensing Letters. 2017;14: 1101-1105. DOI: 10.1109/LGRS. 2017.2698450

[22] Ma W, Wen Z, Wu Y, Jiao L, Gong M, Zheng Y, et al. Remote sensing image registration with modified SIFT and enhanced feature matching. IEEE

Geoscience and Remote Sensing Letters. 2017;14:3-7. DOI: 10.1109/LGRS. 2016.2600858

[23] Xiang Y, Wang F, You H. OS-SIFT: A robust SIFT-like algorithm for highresolution optical-to-SAR image registration in suburban areas. IEEE Transactions on Geoscience and Remote Sensing. 2018;56:3078-3090. DOI: 10.1109/TGRS.2018.2790483

[24] Teke M, Temizel A. Multi-spectral satellite image registration using scalerestricted SURF. In: Proceedings of International Conference on Pattern Recognition (ICPR'10); 23–26 August 2010; Istanbul, Turkey. pp. 2310-2313

[25] Nguyen H-G, Fablet R, Ehrhold A, Boucher J-M. Keypoint-based analysis of sonar images: Applications to seabed recognition. IEEE Transactions on Geoscience and Remote Sensing. 2012; 50:1171-1184. DOI: 10.1109/ TGRS.2011.2165848

[26] Liu R, Wang Y. SAR image matching based on speeded up robust feature. In: Proceedings of the WRI Global Congress on Intelligent System (GCIS'09); 19–21 May, 2009; Xiamen, China. pp. 518-522

[27] Lee WK, Kim AL. An efficient automatic geo-registration technique for high resolution spaceborne SAR image fusion. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium (IGASS'11); 24–29 July, 2011; Vancouver, Canada. pp. 3566-3569

[28] Zhang H, Ni W, Yan W, Wu J, Li S. Robust SAR image registration based on edge matching and refined coherent point drift. IEEE Geoscience and Remote Sensing Letters. 2015;12: 2115-2119. DOI: 10.1109/LGRS. 2015.2451396

[29] Li D, Zhang Y. On the appropriate feature for general SAR image registration. Proceedings of SPIE. 2012; 8536:85360X. DOI: 10.1117/12.970520

[30] Fischler MA, Bolles RC. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM. 1981;24: 381-395. DOI: 10.1145/358669.358692

[31] Zuliani M. Computational methods for automatic image registration [thesis]. Santa Barbara, U.S.: University of California at Santa Barbara; 2007

[32] Massart DL, Kaufman L, Rousseeuw PJ, Leroy A. Least median of squares: A robust method for outlier and model error detection in regression and calibration. Analytica Chimica Acta. 1986;187:171-179. DOI: 10.1016/ S0003-2670(00)82910-4

[33] Rousseeuw PJ, Van Driessen K. Computing LTS regression for large data sets. Data Mining and Knowledge Discovery. 2006;12:29-45. DOI: 10.1007/s10618-005-0024-4

[34] Zhang Z, Deriche R, Faugeras O, Luong QT. A robust technique for matching two uncalibrated images through the recovery of the unknown epipolar geometry. Artificial Intelligence. 1995;78:87-119. DOI: 10.1016/0004-3702(95)00022-4

[35] Li D, Zhang Y. The appropriate parameter retrieval algorithm for feature-based SAR image registration. Proceedings of SPIE. 2012;8536:85360Y. DOI: 10.1117/12.970522

[36] Brown LG. A survey of image registration techniques. ACM Computing Surveys. 1992;24:325-376. DOI: 10.1145/146370.146374

[37] Zitova B, Flusser J. Image registration methods: A survey. Image and Vision Computing. 2003;21:

977-1000. DOI: 10.1016/S0262-8856 (03)00137-9

correlation and split-bandwidth

pp. 143-177

interferometry for wideband and Deltak SAR systems. IEEE Geoscience and Remote Sensing Letters. 2005;2:151-155. DOI: 10.1109/LGRS.2004.843203

DOI: http://dx.doi.org/10.5772/intechopen.81665

54:723-743. DOI: 10.1109/TGRS.

[53] Li D, Zhang Y. Adaptive modelbased classification of PolSAR data. IEEE Transactions on Geoscience and Remote Sensing. 2018;56:1-16. DOI: 10.1109/TGRS.2018.2845944

2015.2464113

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm

[46] Lee JS, Pottier E. Polarimetric Radar Imaging: From Basics to Applications. Boca Raton, U.S.: CRC Press; 2009.

[47] Li D, Zhang Y. A rigorous SAR epipolar geometry modeling and application to 3D target reconstruction. IEEE Selected Topics in Applied Earth Observations and Remote Sensing. 2013;

[48] Nitti DO, Hanssen RF, Refice A, Bovenga F, Nutricato R. Impact of DEM-assisted coregistration on highresolution SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2011;49:1127-1143. DOI: 10.1109/TGRS.2010.2074204

[49] Gabriel AK, Goldstein RM. Crossed orbit interferometry: Theory and experimental results from SIR-B. International Journal of Remote Sensing.

1988;9:857-872. DOI: 10. 1080/

2005. pp. 1-157. DOI: 10.1002/

[50] Rousseeuw PJ, Leroy A. Robust Regression and Outlier Detection. Hoboken, U.S.: John Wiley & Sons, Inc.;

[51] Sweet AD, Dubbert DF, Doerry AW, Sloan GR, Dee Gutierrez V. A portfolio of fine resolution Ku-band MiniSAR images. In: Proceedings of SPIE Defense and Security Symposium; 17–21 April, 2006; Orlando, U.S.

[52] Li D, Zhang Y. Unified Huynen phenomenological decomposition of radar targets and its classification applications. IEEE Transactions on Geoscience and Remote Sensing. 2016;

01431168808954901

pp. 621002-621006

45

0471725382

6:2316-2323. DOI: 10.1109/ JSTARS.2013.2249575

[38] Dawn S, Saxena V, Sharma B. Remote sensing image registration techniques: A survey. In: Proceedings of International Conference on Image and Signal Processing (ICISP'10); 30 June–2 July, 2010; Trois-Rivieres, Canada. pp. 103-112

[39] Mikolajczyk K, Schmid C. An affine invariant interest point detector. In: Proceedings of European Conference on Computer Vision (ECCV'02); 28–31 May, 2002; Copenhagen, Denmark. pp. 128-142

[40] Rufino G, Moccia A, Esposito S. DEM generation by means of ERS tandem data. IEEE Transactions on Geoscience and Remote Sensing. 1998; 36:1905-1912. DOI: 10.1109/36. 729362

[41] Li FK, Goldstein RM. Studies of multibaseline spaceborne interferometric synthetic aperture radars. IEEE Transactions on Geoscience and Remote Sensing. 1990; 28:88-97. DOI: 10.1109/36.45749

[42] Brown M, Lowe DG. Invariant features from interest point groups. In: Proceedings of the British Machine Vision Conference (BMVC'02); 2–5 September, 2002; Cardiff, U.K. pp. 253-262

[43] Li D, Zhang Y. A novel approach for the registration of weak affine images. Pattern Recognition Letters. 2012;33: 1647-1655. DOI: 10.1016/j. patrec.2012.04.009

[44] Li Z, Bethel J. Image coregistration in SAR interferometry. In: Archives of International Society got Photogrammetry and Remote Sensing (ISPRS'08); 3–11 July 2008; Beijing, China. pp. 433-438

[45] Bamler R, Eineder M. Accuracy of differential shift estimation by

On Feature-Based SAR Image Registration: Appropriate Feature and Retrieval Algorithm DOI: http://dx.doi.org/10.5772/intechopen.81665

correlation and split-bandwidth interferometry for wideband and Deltak SAR systems. IEEE Geoscience and Remote Sensing Letters. 2005;2:151-155. DOI: 10.1109/LGRS.2004.843203

[29] Li D, Zhang Y. On the appropriate

977-1000. DOI: 10.1016/S0262-8856

[38] Dawn S, Saxena V, Sharma B. Remote sensing image registration techniques: A survey. In: Proceedings of International Conference on Image and Signal Processing (ICISP'10); 30 June–2 July, 2010; Trois-Rivieres, Canada.

[39] Mikolajczyk K, Schmid C. An affine invariant interest point detector. In: Proceedings of European Conference on Computer Vision (ECCV'02); 28–31 May, 2002; Copenhagen, Denmark.

[40] Rufino G, Moccia A, Esposito S. DEM generation by means of ERS tandem data. IEEE Transactions on Geoscience and Remote Sensing. 1998; 36:1905-1912. DOI: 10.1109/36. 729362

[41] Li FK, Goldstein RM. Studies of

interferometric synthetic aperture radars. IEEE Transactions on

[42] Brown M, Lowe DG. Invariant features from interest point groups. In: Proceedings of the British Machine Vision Conference (BMVC'02); 2–5 September, 2002; Cardiff, U.K.

Geoscience and Remote Sensing. 1990; 28:88-97. DOI: 10.1109/36.45749

[43] Li D, Zhang Y. A novel approach for the registration of weak affine images. Pattern Recognition Letters. 2012;33:

[44] Li Z, Bethel J. Image coregistration in SAR interferometry. In: Archives of

Photogrammetry and Remote Sensing (ISPRS'08); 3–11 July 2008; Beijing,

[45] Bamler R, Eineder M. Accuracy of

differential shift estimation by

1647-1655. DOI: 10.1016/j. patrec.2012.04.009

International Society got

China. pp. 433-438

multibaseline spaceborne

(03)00137-9

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

pp. 103-112

pp. 128-142

pp. 253-262

registration. Proceedings of SPIE. 2012; 8536:85360X. DOI: 10.1117/12.970520

[30] Fischler MA, Bolles RC. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM. 1981;24: 381-395. DOI: 10.1145/358669.358692

[31] Zuliani M. Computational methods for automatic image registration [thesis]. Santa Barbara, U.S.: University of California at Santa Barbara; 2007

[32] Massart DL, Kaufman L, Rousseeuw PJ, Leroy A. Least median of squares: A robust method for outlier and model error detection in regression and calibration. Analytica Chimica Acta. 1986;187:171-179. DOI: 10.1016/ S0003-2670(00)82910-4

[33] Rousseeuw PJ, Van Driessen K. Computing LTS regression for large data sets. Data Mining and Knowledge Discovery. 2006;12:29-45. DOI: 10.1007/s10618-005-0024-4

[34] Zhang Z, Deriche R, Faugeras O, Luong QT. A robust technique for matching two uncalibrated images through the recovery of the unknown

[35] Li D, Zhang Y. The appropriate parameter retrieval algorithm for feature-based SAR image registration. Proceedings of SPIE. 2012;8536:85360Y.

[36] Brown LG. A survey of image registration techniques. ACM

DOI: 10.1145/146370.146374

[37] Zitova B, Flusser J. Image

44

Computing Surveys. 1992;24:325-376.

registration methods: A survey. Image and Vision Computing. 2003;21:

epipolar geometry. Artificial Intelligence. 1995;78:87-119. DOI: 10.1016/0004-3702(95)00022-4

DOI: 10.1117/12.970522

feature for general SAR image

[46] Lee JS, Pottier E. Polarimetric Radar Imaging: From Basics to Applications. Boca Raton, U.S.: CRC Press; 2009. pp. 143-177

[47] Li D, Zhang Y. A rigorous SAR epipolar geometry modeling and application to 3D target reconstruction. IEEE Selected Topics in Applied Earth Observations and Remote Sensing. 2013; 6:2316-2323. DOI: 10.1109/ JSTARS.2013.2249575

[48] Nitti DO, Hanssen RF, Refice A, Bovenga F, Nutricato R. Impact of DEM-assisted coregistration on highresolution SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2011;49:1127-1143. DOI: 10.1109/TGRS.2010.2074204

[49] Gabriel AK, Goldstein RM. Crossed orbit interferometry: Theory and experimental results from SIR-B. International Journal of Remote Sensing. 1988;9:857-872. DOI: 10. 1080/ 01431168808954901

[50] Rousseeuw PJ, Leroy A. Robust Regression and Outlier Detection. Hoboken, U.S.: John Wiley & Sons, Inc.; 2005. pp. 1-157. DOI: 10.1002/ 0471725382

[51] Sweet AD, Dubbert DF, Doerry AW, Sloan GR, Dee Gutierrez V. A portfolio of fine resolution Ku-band MiniSAR images. In: Proceedings of SPIE Defense and Security Symposium; 17–21 April, 2006; Orlando, U.S. pp. 621002-621006

[52] Li D, Zhang Y. Unified Huynen phenomenological decomposition of radar targets and its classification applications. IEEE Transactions on Geoscience and Remote Sensing. 2016; 54:723-743. DOI: 10.1109/TGRS. 2015.2464113

[53] Li D, Zhang Y. Adaptive modelbased classification of PolSAR data. IEEE Transactions on Geoscience and Remote Sensing. 2018;56:1-16. DOI: 10.1109/TGRS.2018.2845944

Chapter 3

Ryo Natsuaki

1. Introduction

47

dence of the quay wall with 3 cm error.

Abstract

L-Band SAR Disaster Monitoring

Synthetic aperture radar (SAR) has become a major tool for disaster monitoring.

Its all-weather capability enables us to monitor the affected area soon after the event happens. Since the first launch of spaceborne SAR, its amplitude images have been widely used for disaster observations. Nowadays, an accurate orbit control and scheduled frequent observations enable us to perform interferometric analysis of SAR (InSAR) and the use of interferometric coherence. Especially for L-band SAR, its long-lasting temporal coherence is an advantage to perform precise interferometric coherence analysis. In addition, recent high resolution SAR images are found to be useful for observing relatively small targets, e.g., individual buildings and facilities. In this chapter, we present basic theory of SAR observation, interferometric coherence analysis for the disaster monitoring, and its examples for the harbor facilities. In the actual case, DInSAR measurement could measure the subsi-

Keywords: synthetic aperture radar (SAR), interferometry, interferometric

SAR, its long-lasting temporal coherence enables us to perform precise multitemporal interferometric coherence analysis [14]. Polarimetric analysis (PolSAR) has been also proposed for the damage detection using scattering mechanism analysis [15]. In PolSAR mode, SAR transmits both horizontal and vertical polarized waves and receives their co- and cross-polarized signals to see the

In the last decade, Interferometric Synthetic Aperture Radar (InSAR) has widely spread for measuring ground deformations caused by disasters, for example, earthquakes, volcanic eruptions, or subsidence [1, 2]. It can measure several centimeters of deformation with one pair of SAR images. The accuracy can be increased to several millimeters per year by applying time-series analysis [3, 4]. Compared with traditional optical or amplitude-based SAR analyses, e.g., [5], the advantages of InSAR-based monitoring are, for example, its sensitivity for the deformation and all-weather availability [6–8]. It is effective in the detection of various deformations caused by disasters such as earthquakes [9, 10], volcanic eruptions [11], storms [12], and human disaster [13]. Especially for the long wavelength SAR, i.e., L-band

coherence, disaster monitoring, infrastructure monitoring

for Harbor Facilities Using

Interferometric Analysis

## Chapter 3

## L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

Ryo Natsuaki

## Abstract

Synthetic aperture radar (SAR) has become a major tool for disaster monitoring. Its all-weather capability enables us to monitor the affected area soon after the event happens. Since the first launch of spaceborne SAR, its amplitude images have been widely used for disaster observations. Nowadays, an accurate orbit control and scheduled frequent observations enable us to perform interferometric analysis of SAR (InSAR) and the use of interferometric coherence. Especially for L-band SAR, its long-lasting temporal coherence is an advantage to perform precise interferometric coherence analysis. In addition, recent high resolution SAR images are found to be useful for observing relatively small targets, e.g., individual buildings and facilities. In this chapter, we present basic theory of SAR observation, interferometric coherence analysis for the disaster monitoring, and its examples for the harbor facilities. In the actual case, DInSAR measurement could measure the subsidence of the quay wall with 3 cm error.

Keywords: synthetic aperture radar (SAR), interferometry, interferometric coherence, disaster monitoring, infrastructure monitoring

## 1. Introduction

In the last decade, Interferometric Synthetic Aperture Radar (InSAR) has widely spread for measuring ground deformations caused by disasters, for example, earthquakes, volcanic eruptions, or subsidence [1, 2]. It can measure several centimeters of deformation with one pair of SAR images. The accuracy can be increased to several millimeters per year by applying time-series analysis [3, 4]. Compared with traditional optical or amplitude-based SAR analyses, e.g., [5], the advantages of InSAR-based monitoring are, for example, its sensitivity for the deformation and all-weather availability [6–8]. It is effective in the detection of various deformations caused by disasters such as earthquakes [9, 10], volcanic eruptions [11], storms [12], and human disaster [13]. Especially for the long wavelength SAR, i.e., L-band SAR, its long-lasting temporal coherence enables us to perform precise multitemporal interferometric coherence analysis [14]. Polarimetric analysis (PolSAR) has been also proposed for the damage detection using scattering mechanism analysis [15]. In PolSAR mode, SAR transmits both horizontal and vertical polarized waves and receives their co- and cross-polarized signals to see the scattering mechanism of targets. The collapsed buildings show different scattering mechanisms when they are compared with standing buildings. The most important examples are derived from 2011 off the pacific coast of Tohoku Earthquake [16–19]. Current problem for this method in the operational SARs is that there is less acquisition for full polarimetric mode, narrower swath width, and less spatial resolution caused by the operational limit of the platforms.

2. Synthetic aperture radar and interferometric analysis

tion of InSAR and coherence analysis.

DOI: http://dx.doi.org/10.5772/intechopen.81465

2.1 Synthetic aperture radar

ground targets are fixed [27].

be analyzed by interferometric analysis.

2.2 Differential interferometric SAR

and tropospheric delay information [28].

49

In this section, we briefly explain the system of synthetic aperture radar (SAR)

The larger antenna diameter derives the higher spatial resolution of radar systems. However, some platforms such as aircraft and satellites cannot deploy a sufficient size of the antenna because of their payload limitations. SAR solves this problem by moving itself and synthesizes the received signals by assuming that the

If a SAR can use wide bandwidth, e.g., 1 GHz, it can achieve approximately 0.25 m of spatial resolution in the range direction. A typical high resolution SAR achieves 3–5 m. In the azimuth direction, a pulse repetition frequency (PRF) and aperture length are the large factors. The amplitude of a pixel of SAR image depends on the backscattering coefficient, and the phase depends on the distance between SAR and scatter. The phase information is difficult to handle because the wavelength is too short to measure the ground directly. On the other hand, the phase contains topographic, deformation, and the other valuable information. Those can

Figure 1 presents a schematic diagram of InSAR analysis. A SAR image contains amplitude and phase information, in other words, complex-valued information and thus is called Single Look Complex (SLC) image. When we observe the same place from the same orbit multiply and multiply one SLC image (master) and another complex conjugated SLC image (slave), we can make an interferogram. The phase value of the interferogram, φ, is the phase difference between the master and slave. A SAR interferometric phase contains topographic, deformation, ionospheric delay,

In this chapter, we consider that the interferogram, φ, consists of the topographic, φtopo, and deformation, φdefo, components and ignore the others. The topographic component can be estimated by calculating the relationship between the known heights H acquired from a known topographic map as shown in Eq. (1).

> <sup>φ</sup>topo <sup>¼</sup> <sup>4</sup>πBCT cos <sup>θ</sup> � <sup>γ</sup>CT ð Þ λRm sin θ

where λ is the wavelength of SAR, BCT cos(θ-γCT) is the perpendicular baseline of the two observations, θ is the incidence angle, and Rm is the slant range distance. Therefore, we can subtract φtopo from the interferogram and measure the deformation component. If deformation occurs between two observations, the deformation phase value, φdefo, corresponds to the shrink or extension in the line-of-sight

H (1)

and its interferometric analysis (InSAR), including coherence analysis. Three monitoring methods are shown here. Firstly, wreckages and inundated area detection using amplitude information are described. Next, DInSAR-based displacement detection of the ground is explained. Finally, assessment for the damaged buildings from interferometric coherence is introduced. The descriptions are especially supposed for damage detection in harbor facilities, which is a combina-

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

Another merit of spaceborne observations is that they have a wider observation swath than that of airborne observations, resulting in faster measurement over a wide area. However, the area of deformation is assumed to be larger than hundreds of meters in the world of InSAR. If its spatial resolution increases, it can be applied to smaller targets' deformation, e.g., the disaster monitoring of harbor facilities smaller than 100 m [20], in addition to the existing change detection methods, e.g., [21, 22].

Currently, operational SAR satellites aim wide swath or high resolution. ALOS-2, COSMO-SkyMed, RADARSAT-2, and TerraSAR/TanDEM-X aim higher resolution (<5 m) with relatively narrow swath width (<50 km), while Sentinel-1 aims wider swath width (200 km) with lower resolution (>20 m). Observation with higher resolution can achieve precise texture of the ground. One can analyze individual buildings with such a high resolution, while preceding researches mostly aim to evaluate in the size of a city block [23, 24]. On the other hand, wider observation swath is required for frequent and scheduled global observation using small number of satellites. That is, the higher frequent acquisition enables us to analyze the region of interest (RoI) without making any conflict with other observation requirements. The frequent observation is a requirement not only for time-series analysis but also for disaster monitoring that the users must observe the affected area as soon as possible. In the next decade, wide swath and high resolution are going to be combined, and the Earth will be observed weekly or bi-weekly with higher than 10 m resolution by SAR satellites such as ALOS-4, NISAR, Sentinel-1 NG, and TanDEM-L.

In such an era, disaster monitoring with SAR data using interferometric analysis becomes more useful [25]. In addition to the traditional amplitude-based change detection, centimeter-order deformation detection and interferometric coherencebased damage assessment will be more operational. One can acquire a delineation map over dozens of square kilometers for the affected area with a few meters resolution several hours after the observation, which cannot be achieved with ground/airborne surveys.

Monitoring harbor facilities plays an important role in the recovery phase in the disaster, because maritime traffic is a backbone of the logistics. For example, a heavy storm may have damaged the seawalls and piers. However, it is difficult to assess the stability of them soon after the event by humans because the ocean is still heavy. A catastrophic earthquake and tsunamis may have damaged a number of harbors simultaneously. In such a case, the authorities have to assess the damage of their facilities and decide whether to rearrange the route. SAR can quickly observe the affected area remotely on behalf of the risky direct observation by humans. This is the reason why SAR can play an important role in the rescue and recovery phase of the disaster. This chapter thinks of it.

In this chapter, we firstly present a fundamental theory for the interferometric analysis of SAR. It includes the basis of differential InSAR (DInSAR) and interferometric coherence analysis. Next, we describe a basic scheme of harbor monitoring for disaster monitoring. Finally, we show several examples in the real case, including the latest L-band SAR satellite Advanced Land Observing Satellite-2 (ALOS-2 or DAICHI-2) [26].

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

### 2. Synthetic aperture radar and interferometric analysis

In this section, we briefly explain the system of synthetic aperture radar (SAR) and its interferometric analysis (InSAR), including coherence analysis. Three monitoring methods are shown here. Firstly, wreckages and inundated area detection using amplitude information are described. Next, DInSAR-based displacement detection of the ground is explained. Finally, assessment for the damaged buildings from interferometric coherence is introduced. The descriptions are especially supposed for damage detection in harbor facilities, which is a combination of InSAR and coherence analysis.

#### 2.1 Synthetic aperture radar

scattering mechanism of targets. The collapsed buildings show different scattering mechanisms when they are compared with standing buildings. The most important examples are derived from 2011 off the pacific coast of Tohoku Earthquake [16–19]. Current problem for this method in the operational SARs is that there is less acquisition for full polarimetric mode, narrower swath width, and less spatial

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Another merit of spaceborne observations is that they have a wider observation swath than that of airborne observations, resulting in faster measurement over a wide area. However, the area of deformation is assumed to be larger than hundreds of meters in the world of InSAR. If its spatial resolution increases, it can be applied to smaller targets' deformation, e.g., the disaster monitoring of harbor facilities smaller than 100 m [20], in addition to the existing change detection

Currently, operational SAR satellites aim wide swath or high resolution. ALOS-2, COSMO-SkyMed, RADARSAT-2, and TerraSAR/TanDEM-X aim higher resolution (<5 m) with relatively narrow swath width (<50 km), while Sentinel-1 aims wider swath width (200 km) with lower resolution (>20 m). Observation with higher resolution can achieve precise texture of the ground. One can analyze individual buildings with such a high resolution, while preceding researches mostly aim to evaluate in the size of a city block [23, 24]. On the other hand, wider observation swath is required for frequent and scheduled global observation using small number of satellites. That is, the higher frequent acquisition enables us to analyze the region of interest (RoI) without making any conflict with other observation requirements. The frequent observation is a requirement not only for time-series analysis but also for disaster monitoring that the users must observe the affected area as soon as possible. In the next decade, wide swath and high resolution are going to be combined, and the Earth will be observed weekly or bi-weekly with higher than 10 m resolution by SAR satellites such as ALOS-4, NISAR, Sentinel-1 NG, and

In such an era, disaster monitoring with SAR data using interferometric analysis becomes more useful [25]. In addition to the traditional amplitude-based change detection, centimeter-order deformation detection and interferometric coherencebased damage assessment will be more operational. One can acquire a delineation map over dozens of square kilometers for the affected area with a few meters resolution several hours after the observation, which cannot be achieved with

Monitoring harbor facilities plays an important role in the recovery phase in the

In this chapter, we firstly present a fundamental theory for the interferometric analysis of SAR. It includes the basis of differential InSAR (DInSAR) and interferometric coherence analysis. Next, we describe a basic scheme of harbor monitoring for disaster monitoring. Finally, we show several examples in the real case, including the latest L-band SAR satellite Advanced Land Observing Satellite-2 (ALOS-2 or

disaster, because maritime traffic is a backbone of the logistics. For example, a heavy storm may have damaged the seawalls and piers. However, it is difficult to assess the stability of them soon after the event by humans because the ocean is still heavy. A catastrophic earthquake and tsunamis may have damaged a number of harbors simultaneously. In such a case, the authorities have to assess the damage of their facilities and decide whether to rearrange the route. SAR can quickly observe the affected area remotely on behalf of the risky direct observation by humans. This is the reason why SAR can play an important role in the rescue and recovery phase

resolution caused by the operational limit of the platforms.

methods, e.g., [21, 22].

TanDEM-L.

ground/airborne surveys.

DAICHI-2) [26].

48

of the disaster. This chapter thinks of it.

The larger antenna diameter derives the higher spatial resolution of radar systems. However, some platforms such as aircraft and satellites cannot deploy a sufficient size of the antenna because of their payload limitations. SAR solves this problem by moving itself and synthesizes the received signals by assuming that the ground targets are fixed [27].

If a SAR can use wide bandwidth, e.g., 1 GHz, it can achieve approximately 0.25 m of spatial resolution in the range direction. A typical high resolution SAR achieves 3–5 m. In the azimuth direction, a pulse repetition frequency (PRF) and aperture length are the large factors. The amplitude of a pixel of SAR image depends on the backscattering coefficient, and the phase depends on the distance between SAR and scatter. The phase information is difficult to handle because the wavelength is too short to measure the ground directly. On the other hand, the phase contains topographic, deformation, and the other valuable information. Those can be analyzed by interferometric analysis.

#### 2.2 Differential interferometric SAR

Figure 1 presents a schematic diagram of InSAR analysis. A SAR image contains amplitude and phase information, in other words, complex-valued information and thus is called Single Look Complex (SLC) image. When we observe the same place from the same orbit multiply and multiply one SLC image (master) and another complex conjugated SLC image (slave), we can make an interferogram. The phase value of the interferogram, φ, is the phase difference between the master and slave. A SAR interferometric phase contains topographic, deformation, ionospheric delay, and tropospheric delay information [28].

In this chapter, we consider that the interferogram, φ, consists of the topographic, φtopo, and deformation, φdefo, components and ignore the others. The topographic component can be estimated by calculating the relationship between the known heights H acquired from a known topographic map as shown in Eq. (1).

$$\rho\_{\text{topo}} = \frac{4\pi B\_{CT}\cos\left(\theta - \chi\_{CT}\right)}{\lambda R\_m \sin\theta} H \tag{1}$$

where λ is the wavelength of SAR, BCT cos(θ-γCT) is the perpendicular baseline of the two observations, θ is the incidence angle, and Rm is the slant range distance. Therefore, we can subtract φtopo from the interferogram and measure the deformation component. If deformation occurs between two observations, the deformation phase value, φdefo, corresponds to the shrink or extension in the line-of-sight

Figure 1. Schematic diagram of InSAR analysis.

distance between the satellite and ground targets. If we denote the change in the line-of-sight distance as ΔR, the phase value can be calculated with the wave length λ using Eq. (2).

$$
\rho\_{\text{defo}} = \frac{4\pi}{\lambda} \Delta R \tag{2}
$$

methods have been proposed. The Goldstein-Werner filter [29] is the famous low pass filter in the frequency domain. Probability estimation methods such as Markov random field model [30] and Bayesian estimation [31] have been proposed too. Nonlocal filter is widely used for its robustness [32]. Robust unwrapping methods have also been proposed. Branch-cut technique [33] tries to find the minimum cost to cancel the SPs by connecting opposite rotation side ones. Least square methods [34, 35] use Fourier transformation to distinguish steep slope from high frequency noise. The singularity spreading technique [36] is a newly developed method, which simply cancels residues by adding opposite direction to send residue to the other residues. In this chapter, we applied Markov random field model [30] filter and a

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

Interferometric coherence represents the uniformity of the interferometric pair

of the SAR images [37, 38]. Interferometric coherence becomes high when the Master and Slave images are close to each other, while it decreases when two are completely different. The coherence value is calculated from the cross-correlation and autocorrelation between the two observations as shown in Eq. (3). When the ground targets are damaged or collapsed by disasters or human activities, the identical position reflects radio waves differently when we compare the pre- and post-event SLCs. In this case, the interferometric phase value contains no information, and the signal in master and slave SLC has no correlation. If the ground surface has been changed by the disaster, this effect appears as a large decrease in interfer-

> <sup>γ</sup> <sup>¼</sup> <sup>M</sup>∗<sup>S</sup> � � ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi M∗M q� � ffiffiffiffiffiffiffiffiffiffiffiffi

where M and S represent Master and Slave samples, M∗M represents the complex conjugate multiplication of M, and < > represents the ensemble average of the samples in < >. In short, γ is a normalized cross-correlation of M and S, and thus, it varies from 0 to 1. γ = 1 only happens when M=S and γ = 0 never occurs because of randomness. A large facility has a relatively high (approximately 0.7–0.9) value. Contrarily, bare soil and concrete caissons have low (0.3<) values because they have smooth surface and low reflectivity in radar. An insufficient window size will overestimate the coherence value, while the larger window size will reduce the ground resolution. The window size of the ensemble average in Eq. (3) is 5 � 5 pixels in this chapter. The interferometric coherence largely depends on the surface roughness and temporal stability. If it is too smooth and/or unstable, such as water surface, concrete surface, and highly active region, the radio wave does not return

Coherence γ also depends on the interval of M and S. Though it is stable, the ground surface changes time by time. If master and slave images are acquired in, for example, different years, γ becomes lower. This is called temporal decorrelation. To avoid this effect, it is required to observe the same place frequently. In general, a

When we compare γ of two interferograms, we can detect the effect of disasters. This is called multitemporal coherence analysis; its aim is to detect the damaged part from the change in γ. If we have at least one interferogram prior to the disaster, a pre-event interferogram and an interferogram which is made from pre-event and post-event SLCs, a co-event interferogram, we can compare their coherence values.

S∗S

q� � , (3)

least-square method [35] for phase unwrapping.

2.4 Interferometric coherence analysis

DOI: http://dx.doi.org/10.5772/intechopen.81465

ometric coherence.

to the satellite coherently.

51

lower band SAR has a slower temporal decorrelation.

As shown in the Eq. (2), the range of phase value is limited between �π and π, and thus, we cannot distinguish deformations greater than a quarter wavelength. For example, L-band radar has approximately 24 cm wavelength. When ΔR = 0, +/�12, +/�24… cm, φdefo becomes 0 with indefinite 2nπ. Therefore, we cannot define the deformation if the neighboring pixels have more than a 6 cm line-of-sight difference. Long wavelength has an advantage to measure a large deformation. We can measure the absolute deformation by unwrapping the phase as long as the deformation satisfies the sampling theorem. The robustness of DInSAR analysis for harbor facilities is discussed in [20]. In [20], it is reported that the average error of the 11 observations was 0.1 cm, and its standard deviation was 0.4 cm for ideal case. That is, there was no systematic error of more than 0.1 cm when we apply the averaging filter when it contains a 0.4 cm variation inside the averaging window. At the same time, the average of the standard deviation of every observation was 1.0 cm. That is, a measured deformation with L-band SAR contains a 1.0 cm error.

#### 2.3 Phase unwrapping

In order to calculate the absolute amount of the deformation, phase unwrapping process is required. The exact operation of phase unwrapping is a line integration of the phase values. We can achieve the integration result, which is independent of the integration path, as long as the amount of deformation between all neighboring pixels satisfies the sampling theorem. If there are residual points, i.e. rotational points, the unwrapping results become dependent on the integration path.

To solve this problem, estimating an appropriate phase value (filtering) and finding a specific integration path is required. Various filtering and unwrapping L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

methods have been proposed. The Goldstein-Werner filter [29] is the famous low pass filter in the frequency domain. Probability estimation methods such as Markov random field model [30] and Bayesian estimation [31] have been proposed too. Nonlocal filter is widely used for its robustness [32]. Robust unwrapping methods have also been proposed. Branch-cut technique [33] tries to find the minimum cost to cancel the SPs by connecting opposite rotation side ones. Least square methods [34, 35] use Fourier transformation to distinguish steep slope from high frequency noise. The singularity spreading technique [36] is a newly developed method, which simply cancels residues by adding opposite direction to send residue to the other residues. In this chapter, we applied Markov random field model [30] filter and a least-square method [35] for phase unwrapping.

#### 2.4 Interferometric coherence analysis

Interferometric coherence represents the uniformity of the interferometric pair of the SAR images [37, 38]. Interferometric coherence becomes high when the Master and Slave images are close to each other, while it decreases when two are completely different. The coherence value is calculated from the cross-correlation and autocorrelation between the two observations as shown in Eq. (3). When the ground targets are damaged or collapsed by disasters or human activities, the identical position reflects radio waves differently when we compare the pre- and post-event SLCs. In this case, the interferometric phase value contains no information, and the signal in master and slave SLC has no correlation. If the ground surface has been changed by the disaster, this effect appears as a large decrease in interferometric coherence.

$$\chi = \frac{\langle \mathbf{M} \ast \overline{\mathbf{S}} \rangle}{\sqrt{\langle \mathbf{M} \ast \overline{\mathbf{M}} \rangle} \sqrt{\langle \mathbf{S} \ast \overline{\mathbf{S}} \rangle}},\tag{3}$$

where M and S represent Master and Slave samples, M∗M represents the complex conjugate multiplication of M, and < > represents the ensemble average of the samples in < >. In short, γ is a normalized cross-correlation of M and S, and thus, it varies from 0 to 1. γ = 1 only happens when M=S and γ = 0 never occurs because of randomness. A large facility has a relatively high (approximately 0.7–0.9) value. Contrarily, bare soil and concrete caissons have low (0.3<) values because they have smooth surface and low reflectivity in radar. An insufficient window size will overestimate the coherence value, while the larger window size will reduce the ground resolution. The window size of the ensemble average in Eq. (3) is 5 � 5 pixels in this chapter. The interferometric coherence largely depends on the surface roughness and temporal stability. If it is too smooth and/or unstable, such as water surface, concrete surface, and highly active region, the radio wave does not return to the satellite coherently.

Coherence γ also depends on the interval of M and S. Though it is stable, the ground surface changes time by time. If master and slave images are acquired in, for example, different years, γ becomes lower. This is called temporal decorrelation. To avoid this effect, it is required to observe the same place frequently. In general, a lower band SAR has a slower temporal decorrelation.

When we compare γ of two interferograms, we can detect the effect of disasters. This is called multitemporal coherence analysis; its aim is to detect the damaged part from the change in γ. If we have at least one interferogram prior to the disaster, a pre-event interferogram and an interferogram which is made from pre-event and post-event SLCs, a co-event interferogram, we can compare their coherence values.

distance between the satellite and ground targets. If we denote the change in the line-of-sight distance as ΔR, the phase value can be calculated with the wave length

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

<sup>φ</sup>defo <sup>¼</sup> <sup>4</sup><sup>π</sup> λ

As shown in the Eq. (2), the range of phase value is limited between �π and π, and thus, we cannot distinguish deformations greater than a quarter wavelength. For example, L-band radar has approximately 24 cm wavelength. When ΔR = 0, +/�12, +/�24… cm, φdefo becomes 0 with indefinite 2nπ. Therefore, we cannot define the deformation if the neighboring pixels have more than a 6 cm line-of-sight difference. Long wavelength has an advantage to measure a large deformation. We can measure the absolute deformation by unwrapping the phase as long as the deformation satisfies the sampling theorem. The robustness of DInSAR analysis for harbor facilities is discussed in [20]. In [20], it is reported that the average error of the 11 observations was 0.1 cm, and its standard deviation was 0.4 cm for ideal case. That is, there was no systematic error of more than 0.1 cm when we apply the averaging filter when it contains a 0.4 cm variation inside the averaging window. At the same time, the average of the standard deviation of every observation was 1.0 cm. That is, a measured deformation with L-band SAR contains a 1.0 cm error.

In order to calculate the absolute amount of the deformation, phase unwrapping process is required. The exact operation of phase unwrapping is a line integration of the phase values. We can achieve the integration result, which is independent of the integration path, as long as the amount of deformation between all neighboring pixels satisfies the sampling theorem. If there are residual points, i.e. rotational points, the unwrapping results become dependent on the integration path.

To solve this problem, estimating an appropriate phase value (filtering) and finding a specific integration path is required. Various filtering and unwrapping

ΔR (2)

λ using Eq. (2).

Schematic diagram of InSAR analysis.

Figure 1.

2.3 Phase unwrapping

50

#### Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

This is called multitemporal interferometric coherence change detection. If the coevent interferometric coherence is lower than the pre-event one, though considering temporal decorrelation, it can be regarded that scatterers on the surface have been damaged and/or moved largely. Two definitions for the coherence decrease dγ can be considerable. One is the simple difference or un-normalized coherence decrease (CD) dγun, and the other is the normalized coherence decrease dγnorm, as shown in Eqs. (4) and (5).

$$d\chi\_{un} = \chi\_{pre} - \chi\_{co} \tag{4}$$

$$d\chi\_{norm} = \frac{\chi\_{pre} - \chi\_{co}}{\chi\_{pre} + \chi\_{co}} \tag{5}$$

where the pre-event coherence is γpre and the co-event coherence is γco. Eq. (4) requires relatively large γpre and cannot be applied for low coherency areas, such as vegetated ground. On the other hand, Eq. (5) does not require large γpre, while the temporal decorrelation will suffer in accuracy. In this chapter, Eq. (4) is applied because harbor facilities generally have large γpre in long temporal baseline.

The facilities should be regarded as damaged when dγun exceeds the specific threshold. The threshold is generally defined manually to reduce the effect of temporal and spatial decorrelations. In [39], the authors found that the buildings which are larger than 200m<sup>2</sup> can be evaluated by setting the threshold 0.3. When the buildings exceed the threshold, they were moderately or severely damaged when they are classified by EMS-98 scheme [40]. In this chapter, therefore, threshold for dγun is set to 0.3.

#### 2.5 Scattering mechanisms of harbor facilities

In order to apply those analyses above to the disaster monitoring of harbor facilities, the scattering mechanisms of SAR are briefly described. Numerical models are the same to the other cases; however, it is worth mentioning what the scatterers are in the harbor. Figure 2 shows the schematic image of harbor. In the figure, SAR satellite is observing the harbor from left top of the figure.

Figure 2(a) shows the scattering mechanisms in normal condition. Region A is water, and therefore, backscattering coefficient is very low. Region B is the bare ground, and its brightness depends on its roughness. If the ground is covered by concrete or asphalt, it can be seen dark as same as water. Region C is layover of the facilities, and its surface scattering from the roof can be seen. On the other hand, Region D is a shadow region and Point Z cannot be observed by the satellite. Water, ground, and vertical walls work as smooth surface, and therefore, double-bounce effects are seen in Point X and Y.

Figure 3 presents an example of airborne L-band SAR image and the corresponding spaceborne optical image of the harbor. There are multiple bright targets which exist at the edge of the pier and the buildings. Those are the double-bounce effects. On the other hand, the top of the pier which is covered by asphalt and concrete is mostly dark as the water. The brightness of the rooftop of buildings depends on the structure of them. Those bright scatterers which do not move between two observations have high coherence. Note that ships have no coherence because they move on the water. Other moving facilities such as cars, containers, and cranes do not have high coherence too. On the other hand, stable facilities, e.g., buildings, walls, and vegetation have high coherence in L-band SAR. In short, interferometric analysis is applicable only to the stable bright scatterers.

Figure 2(b) shows a schematic diagram of the traditional amplitude-based SAR analysis. By comparing the pre- and post-disaster observation, the followings can be found. Object P wreckages on the water reflects the radar signal and appears in the post-disaster image. Region Q, an inundated area, will decrease the backscattering coefficients, and therefore, it appears as water region in the postdisaster image. Roads and other smooth surfaces are originally dark in SAR image and therefore may not change the amplitude by flooding. In addition, insufficient amount of subsidence or deformation will not be detected from amplitude image. Region R,

Schematic image for disaster monitoring of harbor facilities using SAR. (a) Scattering mechanism in harbor and

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

DOI: http://dx.doi.org/10.5772/intechopen.81465

change detection based on (b) amplitude and (c) interferometric analysis.

Figure 2.

53

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

Figure 2.

This is called multitemporal interferometric coherence change detection. If the coevent interferometric coherence is lower than the pre-event one, though considering temporal decorrelation, it can be regarded that scatterers on the surface have been damaged and/or moved largely. Two definitions for the coherence decrease dγ can be considerable. One is the simple difference or un-normalized coherence decrease (CD) dγun, and the other is the normalized coherence decrease dγnorm, as

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

<sup>d</sup>γnorm <sup>¼</sup> <sup>γ</sup>pre � <sup>γ</sup>co

where the pre-event coherence is γpre and the co-event coherence is γco. Eq. (4) requires relatively large γpre and cannot be applied for low coherency areas, such as vegetated ground. On the other hand, Eq. (5) does not require large γpre, while the temporal decorrelation will suffer in accuracy. In this chapter, Eq. (4) is applied because harbor facilities generally have large γpre in long temporal baseline.

The facilities should be regarded as damaged when dγun exceeds the specific threshold. The threshold is generally defined manually to reduce the effect of temporal and spatial decorrelations. In [39], the authors found that the buildings which are larger than 200m<sup>2</sup> can be evaluated by setting the threshold 0.3. When the buildings exceed the threshold, they were moderately or severely damaged when they are classified by EMS-98 scheme [40]. In this chapter, therefore, thresh-

In order to apply those analyses above to the disaster monitoring of harbor facilities, the scattering mechanisms of SAR are briefly described. Numerical models are the same to the other cases; however, it is worth mentioning what the scatterers are in the harbor. Figure 2 shows the schematic image of harbor. In the

Figure 2(a) shows the scattering mechanisms in normal condition. Region A is water, and therefore, backscattering coefficient is very low. Region B is the bare ground, and its brightness depends on its roughness. If the ground is covered by concrete or asphalt, it can be seen dark as same as water. Region C is layover of the facilities, and its surface scattering from the roof can be seen. On the other hand, Region D is a shadow region and Point Z cannot be observed by the satellite. Water, ground, and vertical walls work as smooth surface, and therefore, double-bounce

figure, SAR satellite is observing the harbor from left top of the figure.

Figure 3 presents an example of airborne L-band SAR image and the corresponding spaceborne optical image of the harbor. There are multiple bright targets which exist at the edge of the pier and the buildings. Those are the

double-bounce effects. On the other hand, the top of the pier which is covered by asphalt and concrete is mostly dark as the water. The brightness of the rooftop of buildings depends on the structure of them. Those bright scatterers which do not move between two observations have high coherence. Note that ships have no coherence because they move on the water. Other moving facilities such as cars, containers, and cranes do not have high coherence too. On the other hand,

stable facilities, e.g., buildings, walls, and vegetation have high coherence in L-band

SAR. In short, interferometric analysis is applicable only to the stable bright

γpre þ γco

dγun ¼ γpre � γco (4)

(5)

shown in Eqs. (4) and (5).

old for dγun is set to 0.3.

effects are seen in Point X and Y.

scatterers.

52

2.5 Scattering mechanisms of harbor facilities

Schematic image for disaster monitoring of harbor facilities using SAR. (a) Scattering mechanism in harbor and change detection based on (b) amplitude and (c) interferometric analysis.

Figure 2(b) shows a schematic diagram of the traditional amplitude-based SAR analysis. By comparing the pre- and post-disaster observation, the followings can be found. Object P wreckages on the water reflects the radar signal and appears in the post-disaster image. Region Q, an inundated area, will decrease the backscattering coefficients, and therefore, it appears as water region in the postdisaster image. Roads and other smooth surfaces are originally dark in SAR image and therefore may not change the amplitude by flooding. In addition, insufficient amount of subsidence or deformation will not be detected from amplitude image. Region R,

that is, the place where the deformation can be measured has less damage. Region U and V, moderately damaged and collapsed buildings, can be detected by interfero-

In summary, some damages can be detected only by amplitude information,

In a qualitative manner, one can segmentalize the disaster affected harbor facil-

• Nonaffected area can be recognized as high coherence and same amplitude

• Deformation can be measured by DInSAR as long as the surface keeps enough coherence. Note that the phase component of the interferogram is relative value in the line-of-sight direction and not the absolute deformation of neither

• Moderately damaged buildings can be found by the decrease of coherence.

• Severely damaged buildings can be found by both decrease of coherence and

• Inundated areas appear as significant drop of amplitude as well as decrease of

• Wreckages on the water can be found from increase of the amplitude in the

Here, a brief detection scheme is introduced. In the rescue and recovery phase of disaster, mapping an affected area is one of the urgent tasks. The authorities use the delineation map for planning their activities. However, not all the responsible persons are familiar with remote sensing, especially for SAR. Therefore, intuitive

Figure 4 shows an example of the classification flow. The classification scheme consists of five processes. First, the region which amplitude dropped more than 6 dB than pre-disaster data or weaker than the known water region is regarded as under the water and indicated as blue on map. The area which was affected by

subsidence, tidal wave, and/or tsunami will be visualized.

metric coherence analysis too. Its sensitivity is discussed precisely in [39].

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

while the others can be detected only by interferometric analysis. Precise centimeter-order deformation can be measured when the surface keeps enough

coherencies.

areas.

ities as following features.

vertical nor horizontal direction.

DOI: http://dx.doi.org/10.5772/intechopen.81465

increase of amplitude.

3. Damage detection scheme

coherence.

water region.

classification is required.

Decision flow of the quick assessment.

Figure 4.

55

the totally collapsed buildings, can be seen as a brighter scatterer in the postdisaster image, because in general, demolished buildings are more random surface than standing ones. On the other hand, slightly damaged buildings do not show any significant change in the amplitude image. In short, amplitude-based analysis can detect significant difference of the scatterers.

Figure 2(c) shows what can be observed by interferometric analysis. As water and surface has no coherency, wreckages on the water are not visible by interferometric analysis. Inundated Region S shows significant drop of coherence. Region T, deformation of the ground including subsidence or lateral flow, can be seen by interferometric phase and can be measured how large the surface moved. To calculate the absolute amount of the deformation, it requires relatively high coherence,

### L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

that is, the place where the deformation can be measured has less damage. Region U and V, moderately damaged and collapsed buildings, can be detected by interferometric coherence analysis too. Its sensitivity is discussed precisely in [39].

In summary, some damages can be detected only by amplitude information, while the others can be detected only by interferometric analysis. Precise centimeter-order deformation can be measured when the surface keeps enough coherencies.

In a qualitative manner, one can segmentalize the disaster affected harbor facilities as following features.


## 3. Damage detection scheme

Here, a brief detection scheme is introduced. In the rescue and recovery phase of disaster, mapping an affected area is one of the urgent tasks. The authorities use the delineation map for planning their activities. However, not all the responsible persons are familiar with remote sensing, especially for SAR. Therefore, intuitive classification is required.

Figure 4 shows an example of the classification flow. The classification scheme consists of five processes. First, the region which amplitude dropped more than 6 dB than pre-disaster data or weaker than the known water region is regarded as under the water and indicated as blue on map. The area which was affected by subsidence, tidal wave, and/or tsunami will be visualized.

Figure 4. Decision flow of the quick assessment.

the totally collapsed buildings, can be seen as a brighter scatterer in the postdisaster image, because in general, demolished buildings are more random surface than standing ones. On the other hand, slightly damaged buildings do not show any significant change in the amplitude image. In short, amplitude-based analysis can

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Figure 2(c) shows what can be observed by interferometric analysis. As water and surface has no coherency, wreckages on the water are not visible by interferometric analysis. Inundated Region S shows significant drop of coherence. Region T, deformation of the ground including subsidence or lateral flow, can be seen by interferometric phase and can be measured how large the surface moved. To calculate the absolute amount of the deformation, it requires relatively high coherence,

detect significant difference of the scatterers.

Example of (a) SAR and (b) optical observation among harbor facilities.

Figure 3.

54

Second, the region in which amplitude increased more than 6 dB than predisaster data is indicated as yellow on map. Wreckages and totally collapsed buildings will appear here.

Third, the region which coherence dropped more than 0.3 than pre-disaster dataset is regarded as inundated and indicated as red on map. Moderately or severely damaged facilities will be shown in this color. Totally collapsed buildings will be also classified here too.

Fourth, the region in which coherence is higher than 0.6 is regarded as not affected and indicated as green. Showing "safe zone" is demanded for the authorities to decide from where they start their operations.

If the region fulfills both 3 and 4, they do indicate none of them. Some large buildings have higher than 0.9 of coherence in pre-disaster pair, and their coherence may keep higher than 0.6 in co-disaster pair. In such a case, it is difficult to distinguish whether the buildings are damaged or not.

Measurement results for the deformation by DInSAR are presented in the different layer.

## 4. Examples

In this section, we applied the classification scheme to two examples. First is Ishinomaki port, Japan, which was severely affected by the 2011 off the Pacific coast of Tohoku earthquake. The port was observed by ALOS that was operated by JAXA until 2011. The second example is Kumamoto port, Japan, which was slightly affected by 2016 Kumamoto earthquake and observed by the latest ALOS-2. Note that the scheme can be applied to the other disasters such as typhoon and is evaluated in [20].

In Kumamoto port case, we measured the lateral flow with DInSAR. As the area is enough small, we assumed that all phase components consist of topography and deformation and ignored other phase components in the interferogram, such as the ionospheric [41–43] and tropospheric [44, 45] delays.

> Figure 6 shows the closed-up images with optical data of white rectangles, which are marked as X and Y in Figure 5. Figure 6(a) is a closed-up image for the

Damage assessment results for Ishinomaki port, Japan, in 2011 off the Pacific coast of Tohoku earthquake.

demolished, while some of them are remaining. The pier in right hand side has both

Kumamoto port, Kumamoto prefecture, Japan, was hit by the earthquake on April 15, 2016. ALOS-2 had observed the port half a day before the earthquake and observed there again in the next revisit cycle (14-days) on April 29, 2016. There is another observation record from the same orbit on November 14, 2014, and therefore, we can perform the interferometric coherence analysis. The path and frame

Figure 6(b) is a closed-up image and its optical comparison for the Region Y in Figure 5. This part is a breakwater of the port. Soon after the disaster, it is sometimes difficult to approach the offshore facilities. On the other hand, satellite-based SAR can observe them. In this case, tsunami hits the breakwater and some of them are sunk under the water. Wreckages are also found surrounding them. Most damaged buildings were found by interferometric coherence analysis. This is prob-

piers and its comparison in optical images in Region X. Most facilities are

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

DOI: http://dx.doi.org/10.5772/intechopen.81465

ably caused by the orientation, size, and structure of the buildings.

number of the observation are Path 28 and Frame 2930, respectively.

4.2 Kumamoto port, Japan in 2016 Kumamoto earthquake

inundated and wreckage-covered area.

Figure 5.

57

#### 4.1 Ishinomaki port, Japan in 2011 off the Pacific coast of Tohoku earthquake

Ishinomaki port, Miyagi prefecture, Japan, was severely affected by the 2011 off the Pacific coast of Tohoku earthquake on March 11, 2011. A large tsunami hits the port, and almost all facilities were collapsed, damaged, or flushed out. The affected area was observed by ALOS several times. Here, we use the data set of Path 402, Frame 760. The observation dates are April 1, 2011, August 14, 2010, and May 14, 2010. The first two are used for co-event pair, and the latter two are for pre-event pair. Thanks to the L-band SAR's long lasting coherence, 8 months interval pair can be used effectively for the analysis. ALOS has approximately 10 m by 5 m resolution and therefore hardly investigate an identical building in general. However, the harbor facilities in this port were enough large to be distinguished. On the other hand, the deformation itself was too large to be measured by DInSAR. Therefore, we present a delineation map only.

Figure 5 presents the damage assessment results for the Ishinomaki port. As shown in the figure, most part of the coast line of the city was colored in red; the buildings are detected as moderately or severely damaged. In the left side of the figure, there is a large inundated area which is colored in blue. From Figure 5, it is also clear that some buildings on the hills in the north part of the city survived from the earthquake and tsunami.

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

#### Figure 5.

Second, the region in which amplitude increased more than 6 dB than predisaster data is indicated as yellow on map. Wreckages and totally collapsed build-

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Third, the region which coherence dropped more than 0.3 than pre-disaster dataset is regarded as inundated and indicated as red on map. Moderately or severely damaged facilities will be shown in this color. Totally collapsed buildings

Fourth, the region in which coherence is higher than 0.6 is regarded as not affected and indicated as green. Showing "safe zone" is demanded for the authori-

If the region fulfills both 3 and 4, they do indicate none of them. Some large buildings have higher than 0.9 of coherence in pre-disaster pair, and their coherence may keep higher than 0.6 in co-disaster pair. In such a case, it is difficult to

Measurement results for the deformation by DInSAR are presented in the dif-

In this section, we applied the classification scheme to two examples. First is Ishinomaki port, Japan, which was severely affected by the 2011 off the Pacific coast of Tohoku earthquake. The port was observed by ALOS that was operated by JAXA until 2011. The second example is Kumamoto port, Japan, which was slightly affected by 2016 Kumamoto earthquake and observed by the latest ALOS-2. Note that the scheme can be applied to the other disasters such as typhoon and is

In Kumamoto port case, we measured the lateral flow with DInSAR. As the area is enough small, we assumed that all phase components consist of topography and deformation and ignored other phase components in the interferogram, such as the

4.1 Ishinomaki port, Japan in 2011 off the Pacific coast of Tohoku earthquake

Figure 5 presents the damage assessment results for the Ishinomaki port. As shown in the figure, most part of the coast line of the city was colored in red; the buildings are detected as moderately or severely damaged. In the left side of the figure, there is a large inundated area which is colored in blue. From Figure 5, it is also clear that some buildings on the hills in the north part of the city survived from

Ishinomaki port, Miyagi prefecture, Japan, was severely affected by the 2011 off the Pacific coast of Tohoku earthquake on March 11, 2011. A large tsunami hits the port, and almost all facilities were collapsed, damaged, or flushed out. The affected area was observed by ALOS several times. Here, we use the data set of Path 402, Frame 760. The observation dates are April 1, 2011, August 14, 2010, and May 14, 2010. The first two are used for co-event pair, and the latter two are for pre-event pair. Thanks to the L-band SAR's long lasting coherence, 8 months interval pair can be used effectively for the analysis. ALOS has approximately 10 m by 5 m resolution and therefore hardly investigate an identical building in general. However, the harbor facilities in this port were enough large to be distinguished. On the other hand, the deformation itself was too large to be measured by DInSAR. Therefore,

ings will appear here.

ferent layer.

4. Examples

evaluated in [20].

will be also classified here too.

ties to decide from where they start their operations.

distinguish whether the buildings are damaged or not.

ionospheric [41–43] and tropospheric [44, 45] delays.

we present a delineation map only.

the earthquake and tsunami.

56

Damage assessment results for Ishinomaki port, Japan, in 2011 off the Pacific coast of Tohoku earthquake.

Figure 6 shows the closed-up images with optical data of white rectangles, which are marked as X and Y in Figure 5. Figure 6(a) is a closed-up image for the piers and its comparison in optical images in Region X. Most facilities are demolished, while some of them are remaining. The pier in right hand side has both inundated and wreckage-covered area.

Figure 6(b) is a closed-up image and its optical comparison for the Region Y in Figure 5. This part is a breakwater of the port. Soon after the disaster, it is sometimes difficult to approach the offshore facilities. On the other hand, satellite-based SAR can observe them. In this case, tsunami hits the breakwater and some of them are sunk under the water. Wreckages are also found surrounding them. Most damaged buildings were found by interferometric coherence analysis. This is probably caused by the orientation, size, and structure of the buildings.

#### 4.2 Kumamoto port, Japan in 2016 Kumamoto earthquake

Kumamoto port, Kumamoto prefecture, Japan, was hit by the earthquake on April 15, 2016. ALOS-2 had observed the port half a day before the earthquake and observed there again in the next revisit cycle (14-days) on April 29, 2016. There is another observation record from the same orbit on November 14, 2014, and therefore, we can perform the interferometric coherence analysis. The path and frame number of the observation are Path 28 and Frame 2930, respectively.

berthing ships between April 15 and 29. There were almost no damaged

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

damages.

Figure 7.

Interferogram in Kumamoto earthquake.

DOI: http://dx.doi.org/10.5772/intechopen.81465

uled" observation.

59

facilities, and therefore, it is hardly visible to see nongreen part. In Figure 8(b), we show the coherence drop data, red color in Figure 8(a). Now, it is visible that the right hand side of the pier has several damaged facilities. According to the rapid report from the port [46], some roads and facilities received several

In this case, temporal baseline of γpre is almost 17 months. In such a long interval, a SAR, which uses higher frequency (e. g., X- or C-band), cannot achieve enough coherence to compare with γco. These results indicate that L-band SAR may observe the earth from additional incidence angle once in a several years in order to prepare for the disaster. If we have multiple archives from multiple incidence angles, the operator can mauve the satellite to observe the affected area as soon as possible and compare the observation results with the archives. Such operation will greatly help the corresponding authorities because they need not to wait for the next "sched-

Kumamoto port had been under construction to landfill. A lateral flow occurred in the north part of the pier. Figure 8(c) shows the measurement results of the DInSAR. The unwrapped result shows more than 20 cm of line-of-sight displacement. As ALOS-2 observed the port from west of the port, west half of the port moved toward the satellite and the east half moved away from the satellite. On the other hand, the existing parts show only small deformations. For example, the quay wall showed 5–10 cm subsidence by DInSAR measurement. On the other hand, the actual measurement in [46] was 7 cm. Therefore, the error in this case was 3 cm, which is larger than the ideal case in [20] (1 cm). This is caused by, for example, filtering errors, unwrapping errors, or the randomness in the subsidence. In summary, DInSAR could measure the subsidence of the quay wall with several centimeter of error. The measured deformation can be used in the recovery phase of the disaster. As single DInSAR pair can measure the line-of-sight displacement, three dimensional measurements require more than three observations. Especially for the satellite SAR, it is difficult to measure north-south deformation by interferometric analysis because it usually flights polar orbit and can only observe

from east or west. If the deformation is enough large to be detected by co-

registration, pixel-offset method can be applied.

Figure 6. Close up images for Figure 5: (a) Region X and (b) Region Y.

Figure 7 shows an example of the interferometric phase and the position of the port. Each fringe of the interferometric phase represents 12 cm deformation. Fortunately, the port is enough far from the epicenter. Its overall deformation was small enough to continue the operation. Figure 8 presents the delineation map and the analytical results for coherence analysis and DInSAR measurement. Figure 8(a) shows the delineation map. Fortunately, most part of the pier received no damage. Therefore, we could only detect the difference of the

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

Figure 7. Interferogram in Kumamoto earthquake.

berthing ships between April 15 and 29. There were almost no damaged facilities, and therefore, it is hardly visible to see nongreen part. In Figure 8(b), we show the coherence drop data, red color in Figure 8(a). Now, it is visible that the right hand side of the pier has several damaged facilities. According to the rapid report from the port [46], some roads and facilities received several damages.

In this case, temporal baseline of γpre is almost 17 months. In such a long interval, a SAR, which uses higher frequency (e. g., X- or C-band), cannot achieve enough coherence to compare with γco. These results indicate that L-band SAR may observe the earth from additional incidence angle once in a several years in order to prepare for the disaster. If we have multiple archives from multiple incidence angles, the operator can mauve the satellite to observe the affected area as soon as possible and compare the observation results with the archives. Such operation will greatly help the corresponding authorities because they need not to wait for the next "scheduled" observation.

Kumamoto port had been under construction to landfill. A lateral flow occurred in the north part of the pier. Figure 8(c) shows the measurement results of the DInSAR. The unwrapped result shows more than 20 cm of line-of-sight displacement. As ALOS-2 observed the port from west of the port, west half of the port moved toward the satellite and the east half moved away from the satellite. On the other hand, the existing parts show only small deformations. For example, the quay wall showed 5–10 cm subsidence by DInSAR measurement. On the other hand, the actual measurement in [46] was 7 cm. Therefore, the error in this case was 3 cm, which is larger than the ideal case in [20] (1 cm). This is caused by, for example, filtering errors, unwrapping errors, or the randomness in the subsidence. In summary, DInSAR could measure the subsidence of the quay wall with several centimeter of error. The measured deformation can be used in the recovery phase of the disaster. As single DInSAR pair can measure the line-of-sight displacement, three dimensional measurements require more than three observations. Especially for the satellite SAR, it is difficult to measure north-south deformation by interferometric analysis because it usually flights polar orbit and can only observe from east or west. If the deformation is enough large to be detected by coregistration, pixel-offset method can be applied.

Figure 7 shows an example of the interferometric phase and the position of the

port. Each fringe of the interferometric phase represents 12 cm deformation. Fortunately, the port is enough far from the epicenter. Its overall deformation was small enough to continue the operation. Figure 8 presents the delineation map and the analytical results for coherence analysis and DInSAR measurement. Figure 8(a) shows the delineation map. Fortunately, most part of the pier received no damage. Therefore, we could only detect the difference of the

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

Close up images for Figure 5: (a) Region X and (b) Region Y.

Figure 6.

58

5. Conclusion

DOI: http://dx.doi.org/10.5772/intechopen.81465

Acknowledgements

Overseas Research Fellowships.

The author declares no conflict of interest.

Conflict of interest

Author details

Tokyo, Tokyo, Japan

Ryo Natsuaki

61

In this chapter, a rapid damage assessment scheme which is based on SAR interferometric analysis was introduced. With a combination of amplitude analysis, it is able to show an easy-understanding and enough-accurate delineation map. Furthermore, interferometric analysis can provide centimeter-order deformation map. In the example case of Kumamoto earthquake, ALOS-2 detected 5–10 cm subsidence in the quay wall, which was 7 cm in real measurement. These results are highly appreciated by the disaster corresponding authorities. In this chapter, a basic theory is shown. Its accuracy can be easily improved by, for example, machine

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

The ALOS/ALOS-2 original data are copyrighted by MEXT and JAXA, and provided under JAXA 6th ALOS Research Announcement PI No. 3044. The experiments were partially supported by the Council for Science, Technology, and Inno-

Department of Electrical Engineering and Information Systems, The University of

© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium,

\*Address all correspondence to: natsuaki@eis.t.u-tokyo.ac.jp

provided the original work is properly cited.

vation, "Cross-ministerial Strategic Innovation Promotion Program (SIP), Infrastructure Maintenance, Renovation, and Management" (funding agency: NEDO). The maps in this chapter are provided from Geospatial Information Authority under the Ministry of Land, Infrastructure, Transport and Tourism of Japan. Google Inc. provided the aerial images on Google Earth. Dr. Natsuaki is currently with German Aerospace Center as a visiting scientist sponsored by JSPS

learning and data-fusions with the other observations.

#### Figure 8.

Damage assessment results for Kumamoto port, Japan, in 2016 Kumamoto earthquake. (a) Damage assessment map, (b) coherence dropped part only, and (c) deformation map.

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

## 5. Conclusion

In this chapter, a rapid damage assessment scheme which is based on SAR interferometric analysis was introduced. With a combination of amplitude analysis, it is able to show an easy-understanding and enough-accurate delineation map. Furthermore, interferometric analysis can provide centimeter-order deformation map. In the example case of Kumamoto earthquake, ALOS-2 detected 5–10 cm subsidence in the quay wall, which was 7 cm in real measurement. These results are highly appreciated by the disaster corresponding authorities. In this chapter, a basic theory is shown. Its accuracy can be easily improved by, for example, machine learning and data-fusions with the other observations.

## Acknowledgements

The ALOS/ALOS-2 original data are copyrighted by MEXT and JAXA, and provided under JAXA 6th ALOS Research Announcement PI No. 3044. The experiments were partially supported by the Council for Science, Technology, and Innovation, "Cross-ministerial Strategic Innovation Promotion Program (SIP), Infrastructure Maintenance, Renovation, and Management" (funding agency: NEDO). The maps in this chapter are provided from Geospatial Information Authority under the Ministry of Land, Infrastructure, Transport and Tourism of Japan. Google Inc. provided the aerial images on Google Earth. Dr. Natsuaki is currently with German Aerospace Center as a visiting scientist sponsored by JSPS Overseas Research Fellowships.

## Conflict of interest

The author declares no conflict of interest.

## Author details

Ryo Natsuaki Department of Electrical Engineering and Information Systems, The University of Tokyo, Tokyo, Japan

\*Address all correspondence to: natsuaki@eis.t.u-tokyo.ac.jp

© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Figure 8.

60

Damage assessment results for Kumamoto port, Japan, in 2016 Kumamoto earthquake. (a) Damage assessment

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

map, (b) coherence dropped part only, and (c) deformation map.

## References

[1] Boerner WM. Recent advances in extra-wide-band polarimetry, interferometry and polarimetric interferometry in synthetic aperture remote sensing and its applications, radar, sonar navigation. IEE Proceedings. 2003;150(3):113-124

[2] Plank S. Rapid damage assessment by means of multi-temporal SAR - a comprehensive review and outlook to Sentinel-1. Remote Sensing. 2014;6(6): 4870-4906. DOI: 10.3390/rs6064870

[3] Ferretti A, Prati C, Rocca F. Permanent sacatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2001; 39(1):8-20

[4] Ferretti A, Fumagalli A, Novali F, Prati C, Rocca F, Rucci A. A new algorithm for processing interferometric data-stacks: SqueeSAR. IEEE Transactions on Geoscience and Remote Sensing. 2011;49(9):3460-3470

[5] Liu W, Yamazaki F, Adriano B, Mas E, Koshimura S. Development of building height data in Peru from highresolution SAR imagery. Journal of Disaster Research. 2014;9(6): 1042-1049. DOI: 10.20965/jdr.2014. p1042

[6] Matsuoka M, Yamazaki F. Building damage mapping of the 2003 Bam, Iran, earthquake using ENVISAT/ASAR intensity imagery. Earthquake Spectra. 2005;21(S1):S285-S294

[7] Fielding EJ. Surface ruptures and building damage of the 2003 Bam, Iran, earthquake mapped by satellite synthetic aperture radar interferometric correlation. Journal of Geophysical Research. 2005;110(B03302)

[8] Arciniegas GA, Bijker W, Kerle N, Tolpekin VA. Coherence- and amplitude-based analysis of seismogenic damage in Bam, Iran, using ENVISAT ASAR data. IEEE Transactions on Geoscience and Remote Sensing. 2007; 45(6):1571-1581

[16] Sato M, Chen SW, Satake M. Polarimetric SAR analysis of tsunami damage following the march 11, 2011 East Japan earthquake. Proceedings of the IEEE. 2012;100(10):2861-2875

tsunami with L-band SAR fullpolarimetric mode. IEEE Geoscience and Remote Sensing Letters. 2012;9(3):

472-476

6919-6929

2017;12(3):526-535

[21] Matsuoka M, Estrada M.

Development of earthquake-induced building damage estimation model based on ALOS/PALSAR observing the 2007 Peru earthquake. Journal of Disaster Research. 2013;8(2):346-355. DOI: 10.20965/jdr.2013.p0346

[22] Adriano B, Mas E, Koshimura S, Estrada M, Jimenez C. Scenarios of earthquake and tsunami damage probability in Callao region, Peru using tsunami fragility functions. Journal of Disaster Research. 2014;9(6):968-975.

DOI: 10.20965/jdr.2014.p0968

63

[17] Watanabe M, Motohka T, Miyagi Y, Yonezawa C, Shimada M. Analysis of urban areas affected by the 2011 off the pacific coast of Tohoku earthquake and

DOI: http://dx.doi.org/10.5772/intechopen.81465

[23] Yonezawa C, Takeuchi S. Decorrelation of SAR data by urban

damages caused by the 1995 Hyogokennanbu earthquake.

2001;22:1585-1600

2111-2126

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis

International Journal of Remote Sensing.

[24] Matsuoka M, Nojima N. Building damage estimation by integration of seismic intensity information and satellite L-band SAR

imagery. Remote Sensing. 2010;2(9):

[25] Milillo P, Riel B, Minchew B, Yun SH, Simons M, Lundgren P. On the synergistic use of SAR constellations' data exploitation for earth science and natural hazard response. IEEE Journal of

Selected Topics in Applied Earth Observations and Remote Sensing.

[26] Arikawa Y, Saruwatari H, Hatooka Y, Suzuki S. PALSAR-2 launch and early orbit operation result. International Geoscience and Remote Sensing Symposium (IGARSS). 2014;2014:

[27] Cumming IG, Wong FHC. Digital Processing of Synthetic Aperture Radar Data: Algorithms and Implementation. Norwood, MA, USA: Artech House;

[28] Ghiglia DC, Pritt MD. Two-Dimensional Phase Unwrapping: Theory, Algorithms, and Software. John Wiley & Sons, Inc.; 1998:31-58

[29] Goldstein RM, Werner CM. Radar interferogram filtering for geophysical applications. Geophysical Research Letters. 1998;25(21):4035-4038

[30] Yamaki R, Hirose A. Singular unit restoration in interferograms based on complex-valued Markov random field model for phase unwrapping. IEEE Geoscience and Remote Sensing Letters.

2009;6(1):18-22

2016;9(3):1095-1100

3406-3409

2005

[18] Chen SW, Sato M. Tsunami damage investigation of built-up areas using multitemporal spaceborne full polarimetric SAR images. IEEE

Transactions on Geoscience and Remote

[19] Chen SW, Wang XS, Sato M. Urban

[20] Natsuaki R, Anahara T, Kotoura T, Iwatsuka Y, Tomii N, Katayama H, et al. Synthetic aperture radar interferometry for disaster monitoring of harbor facilities. Journal of Disaster Research.

Sensing. 2013;51(4):1985-1997

damage level mapping based on scattering mechanism investigation using fully polarimetric SAR data for the 3.11 East Japan earthquake. IEEE Transactions on Geoscience and Remote Sensing. 2016;54(12):

[9] Natsuaki R, Nagai H, Motohka T, Ohki M, Watanabe M, Thapa RB, et al. SAR interferometry using ALOS-2 PALSAR-2 data for the Mw 7.8 Gorkha Nepal earthquake, Earth, Planets and Space. 2016;68-15:1-13. DOI: 10.1186/ s40623-016-0394-4

[10] Jimenez C, Moggiano N, Mas E, Adriano B, Koshimura S, Fujii Y, et al. Seismic source of 1746 Callao earthquake from tsunami numerical Modeling. Journal of Disaster Research. 2013;8(2):266-273. DOI: 10.20965/ jdr.2013.p0266

[11] Hooper A, Sigmundsson F, Prata F. Remote sensing of volcanic hazards and their precursors. Proceedings of the IEEE. 2012;100(10):2908-2930

[12] Nakmuenwai P, Yamazaki F, Liu W. Multi-temporal correlation method for damage assessment of buildings from high-resolution SAR images of the 2013 typhoon Haiyan. Journal of Disaster Research. 2016;11(3):577-592. DOI: 10.20965/jdr.2016.p0577

[13] Liu W, Yamazaki F, Sasagawa T. Monitoring of the recovery process of the Fukushima Daiichi nuclear power plant from VHR SAR images. Journal of Disaster Research. 2016;11(2):236-245. DOI: 10.20965/jdr.2016.p0236

[14] Wei M, Sandwell DT. Decorrelation of L-band and C-band interferometry over vegetated areas in California. IEEE Transactions on Geoscience and Remote Sensing. 2010;48(7):2942-2952

[15] Yamaguchi Y. Disaster monitoring by fully polarimetric SAR data acquired with ALOS-PALSAR. Proceedings of the IEEE. 2012;100(10):2851-2860

L-Band SAR Disaster Monitoring for Harbor Facilities Using Interferometric Analysis DOI: http://dx.doi.org/10.5772/intechopen.81465

[16] Sato M, Chen SW, Satake M. Polarimetric SAR analysis of tsunami damage following the march 11, 2011 East Japan earthquake. Proceedings of the IEEE. 2012;100(10):2861-2875

References

39(1):8-20

p1042

62

[1] Boerner WM. Recent advances in extra-wide-band polarimetry, interferometry and polarimetric interferometry in synthetic aperture remote sensing and its applications,

damage in Bam, Iran, using ENVISAT ASAR data. IEEE Transactions on Geoscience and Remote Sensing. 2007;

[9] Natsuaki R, Nagai H, Motohka T, Ohki M, Watanabe M, Thapa RB, et al. SAR interferometry using ALOS-2 PALSAR-2 data for the Mw 7.8 Gorkha Nepal earthquake, Earth, Planets and Space. 2016;68-15:1-13. DOI: 10.1186/

[10] Jimenez C, Moggiano N, Mas E, Adriano B, Koshimura S, Fujii Y, et al.

[11] Hooper A, Sigmundsson F, Prata F. Remote sensing of volcanic hazards and their precursors. Proceedings of the IEEE. 2012;100(10):2908-2930

[12] Nakmuenwai P, Yamazaki F, Liu W. Multi-temporal correlation method for damage assessment of buildings from high-resolution SAR images of the 2013 typhoon Haiyan. Journal of Disaster Research. 2016;11(3):577-592. DOI:

[13] Liu W, Yamazaki F, Sasagawa T. Monitoring of the recovery process of the Fukushima Daiichi nuclear power plant from VHR SAR images. Journal of Disaster Research. 2016;11(2):236-245.

[14] Wei M, Sandwell DT. Decorrelation of L-band and C-band interferometry over vegetated areas in California. IEEE Transactions on Geoscience and Remote

[15] Yamaguchi Y. Disaster monitoring by fully polarimetric SAR data acquired with ALOS-PALSAR. Proceedings of the

DOI: 10.20965/jdr.2016.p0236

Sensing. 2010;48(7):2942-2952

IEEE. 2012;100(10):2851-2860

10.20965/jdr.2016.p0577

Seismic source of 1746 Callao earthquake from tsunami numerical Modeling. Journal of Disaster Research. 2013;8(2):266-273. DOI: 10.20965/

45(6):1571-1581

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications,Tsunami…

s40623-016-0394-4

jdr.2013.p0266

[2] Plank S. Rapid damage assessment by means of multi-temporal SAR - a comprehensive review and outlook to Sentinel-1. Remote Sensing. 2014;6(6): 4870-4906. DOI: 10.3390/rs6064870

radar, sonar navigation. IEE Proceedings. 2003;150(3):113-124

[3] Ferretti A, Prati C, Rocca F. Permanent sacatterers in SAR

interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2001;

[4] Ferretti A, Fumagalli A, Novali F, Prati C, Rocca F, Rucci A. A new algorithm for processing interferometric

Transactions on Geoscience and Remote

data-stacks: SqueeSAR. IEEE

Sensing. 2011;49(9):3460-3470

[5] Liu W, Yamazaki F, Adriano B, Mas E, Koshimura S. Development of building height data in Peru from highresolution SAR imagery. Journal of Disaster Research. 2014;9(6): 1042-1049. DOI: 10.20965/jdr.2014.

[6] Matsuoka M, Yamazaki F. Building damage mapping of the 2003 Bam, Iran, earthquake using ENVISAT/ASAR intensity imagery. Earthquake Spectra.

[7] Fielding EJ. Surface ruptures and building damage of the 2003 Bam, Iran,

synthetic aperture radar interferometric correlation. Journal of Geophysical Research. 2005;110(B03302)

[8] Arciniegas GA, Bijker W, Kerle N,

amplitude-based analysis of seismogenic

earthquake mapped by satellite

Tolpekin VA. Coherence- and

2005;21(S1):S285-S294

[17] Watanabe M, Motohka T, Miyagi Y, Yonezawa C, Shimada M. Analysis of urban areas affected by the 2011 off the pacific coast of Tohoku earthquake and tsunami with L-band SAR fullpolarimetric mode. IEEE Geoscience and Remote Sensing Letters. 2012;9(3): 472-476

[18] Chen SW, Sato M. Tsunami damage investigation of built-up areas using multitemporal spaceborne full polarimetric SAR images. IEEE Transactions on Geoscience and Remote Sensing. 2013;51(4):1985-1997

[19] Chen SW, Wang XS, Sato M. Urban damage level mapping based on scattering mechanism investigation using fully polarimetric SAR data for the 3.11 East Japan earthquake. IEEE Transactions on Geoscience and Remote Sensing. 2016;54(12): 6919-6929

[20] Natsuaki R, Anahara T, Kotoura T, Iwatsuka Y, Tomii N, Katayama H, et al. Synthetic aperture radar interferometry for disaster monitoring of harbor facilities. Journal of Disaster Research. 2017;12(3):526-535

[21] Matsuoka M, Estrada M. Development of earthquake-induced building damage estimation model based on ALOS/PALSAR observing the 2007 Peru earthquake. Journal of Disaster Research. 2013;8(2):346-355. DOI: 10.20965/jdr.2013.p0346

[22] Adriano B, Mas E, Koshimura S, Estrada M, Jimenez C. Scenarios of earthquake and tsunami damage probability in Callao region, Peru using tsunami fragility functions. Journal of Disaster Research. 2014;9(6):968-975. DOI: 10.20965/jdr.2014.p0968

[23] Yonezawa C, Takeuchi S. Decorrelation of SAR data by urban damages caused by the 1995 Hyogokennanbu earthquake. International Journal of Remote Sensing. 2001;22:1585-1600

[24] Matsuoka M, Nojima N. Building damage estimation by integration of seismic intensity information and satellite L-band SAR imagery. Remote Sensing. 2010;2(9): 2111-2126

[25] Milillo P, Riel B, Minchew B, Yun SH, Simons M, Lundgren P. On the synergistic use of SAR constellations' data exploitation for earth science and natural hazard response. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2016;9(3):1095-1100

[26] Arikawa Y, Saruwatari H, Hatooka Y, Suzuki S. PALSAR-2 launch and early orbit operation result. International Geoscience and Remote Sensing Symposium (IGARSS). 2014;2014: 3406-3409

[27] Cumming IG, Wong FHC. Digital Processing of Synthetic Aperture Radar Data: Algorithms and Implementation. Norwood, MA, USA: Artech House; 2005

[28] Ghiglia DC, Pritt MD. Two-Dimensional Phase Unwrapping: Theory, Algorithms, and Software. John Wiley & Sons, Inc.; 1998:31-58

[29] Goldstein RM, Werner CM. Radar interferogram filtering for geophysical applications. Geophysical Research Letters. 1998;25(21):4035-4038

[30] Yamaki R, Hirose A. Singular unit restoration in interferograms based on complex-valued Markov random field model for phase unwrapping. IEEE Geoscience and Remote Sensing Letters. 2009;6(1):18-22

[31] Ferraiuolo G, Poggi G. A Bayesian filtering technique for SAR interferometric phase fields. IEEE Transactions on Image Processing. 2004;13(10):1368-1378

[32] Sica F, Cozzolino D, Zhu XX, Verdoliva L, Poggi G. InSAR-BM3D: A nonlocal filter for SAR Interferometric phase restoration. IEEE Transactions on Geoscience and Remote Sensing. 2018; 56:3456-3467

[33] Costantini M. A novel phase unwrapping method based on network programming. IEEE Transactions on Geoscience and Remote Sensing. 1998; 36(3):813-821

[34] Pritt MD, Shipman JS. Least-squares two-dimensional phase unwrapping using FFT's. IEEE Transactions on Geoscience and Remote Sensing. 1994; 32(3):706-708

[35] Suksmono AB, Hirose A. Progressive transform-based phase unwrapping utilizing a recursive structure. IEICE Transactions on Communications. 2006;E89-B(3): 929-936

[36] Yamaki R, Hirose A. Singularityspreading phase unwrapping. IEEE Transactions on Geoscience and Remote Sensing. 2007;45(10):3240-3251

[37] Touzi R, Lopes A, Bruniquel J, Vachon PW. Coherence estimation for SAR imagery. IEEE Transactions on Geoscience and Remote Sensing. 1999; 37(1):135-149. DOI: 10.1109/36.739146

[38] Abdelfattah R, Nicolas JM. Interferometric SAR coherence magnitude estimation using second kind statistics. IEEE Transactions on Geoscience and Remote Sensing. 2003; 44(2):1942-1953

[39] Natsuaki R, Nagai H, Tomii N, Tadono T. Sensitivity and limitation in damage detection for individual buildings using InSAR coherence - a case study in 2016 Kumamoto

earthquakes. Remote Sensing. 2018; 10(2):245

[40] Grunthal G, editor. European Macroseismic Scale 1998. Luxembourg: Centre Europeen de Geodynamique et de Seismologie; 1998

[41] Mayer F, Bamler N, Jakowski R, Fritz T. The potential of low-frequency SAR systems for mapping ionospheric TEC distributions. IEEE Geoscience and Remote Sensing Letters. 2006;3(4): 560-564

[42] Gomba G, Parizzi A, De Zan F, Eineder M, Bamler R. Toward operational compensation of ionospheric effects in SAR interferograms: The split-spectrum method. IEEE Transactions on Geoscience and Remote Sensing. 2015; 54(3):1446-1461. DOI: 10.1109/ TGRS.2015.2481079

[43] Jung HS, Lee WJ. An improvement of ionospheric phase correction by multiple-aperture interferometry. IEEE Transactions on Geoscience and Remote Sensing. 2015;53(9):4952-4960

Section 2

Advanced Image Data

Processing

65

[44] Doin MP, Lasserre C, Peltzer G, Cavalié O, Doubre C. Corrections of stratified tropospheric delays in SAR interferometry: Validation with global atmospheric models. Journal of Applied Geophysics. 2009;69:35-50. DOI: 10.1016/j.jappgeo.2009.03.010

[45] Bekaert DPS, Walters RJ, Wright TJ, Hooper AJ, Parker DJ. Statistical comparison of InSAR tropospheric correction techniques. Remote Sensing of Environment. 2015;170:40-47. DOI: 10.1016/j.rse.2015.08.035

[46] Nodu A. Damage Reports of Ports and Airports, Rapid Communication for 2016 Kumamoto Earthquake, Earthquake Engineering Committee, Japan Society of Civil Engineering. Available from http://committees.jsce. or.jp/eec2/node/76 [in Japanese, Accessed: 2018-08-31]

Section 2
