Introductory Chapter: Automatic Detection of Ice Covers in Airborne Radar Data Using Genetic Algorithm

*Maged Marghany*

## **1. Introduction**

In the polar regions, where extreme weather poses significant challenges to field study, remote sensing from space is one of the most promising technologies being used to observe the environment. The climate, oceans, and terrestrial maritime ecology are all significantly impacted by sea ice, which is one of the most significant markers of changing climate in polar regions. As a consequence, several efforts have been made to keep track of the arctic sea ice.

The dual-core techniques, therefore, for tracking sea ice in the polar areas are the spaceborne radiometer (passive sensor) and scatterometer (active sensor). Ice cover volume, composition, and mobility were each determined using monitoring for the mission. The spaceborne microwave radiometer in particular provides the broadest time series of sea ice cover in the polar regions since 1979, demonstrating a reduction in the average sea ice cover of 0.53 <sup>10</sup><sup>6</sup> km<sup>2</sup> per decade. In this sense, spaceborne synthetic aperture radar (SAR) is a superior option for tracking ice cover from a more comprehensive standpoint because of the major advantages of the high spatial and temporal resolution, polarimetric sensitivity, and configurable imagery modes. While spaceborne SAR can deliver sea ice cover measurements with good resolution at the dimension of 1 km and up to dozens of meters, the radiometer and scatterometer can deliver sea ice accumulation measurements over vast areas with a high resolution of 6.25 km to 12.5 km. In this understanding, Seasat, ERS-1/2, ENVISAT/ASAR, RADARSAT-1/2, TerraSAR-X/TanDEMX, and Sentinel-1 are some of the instances of spaceborne SARs that have proven to be efficient at tracking sea ice details, ice cover and amount, ice classification (such as ice floes, leads, and polynyas), ice movements and meander, icebergs, and ice-wave interactions. Despite techniques for charting sea ice cover and differentiating between ice cover and deep ocean from spaceborne SAR data having already historically been discussed, conventional both Antarctic and Arctic sea ice surveillance has not exploited these kinds of datasets [1–3].

Instinctually, the radar backscatter strength can indeed be used as the basis for the ice-water identification by spaceborne SAR data, because as backscatter of sea ice is usually higher than that of the deeper ocean. This is certainly relevant in the cross-polarization band, which is delicate to quantity scattering whereas the sea surface roughness commonly exhibits surface scattering. On the contrary, once incidence angles fluctuate, the radar backscatter of copolarization (vertical-vertical, VV, or horizontal-horizontal, HH) also varies significantly.

SAR cross-polarization signals, therefore, are demonstrated to be significantly extremely efficacious for identifying glaciers than copolarization SAR images, since the radar backscatter of the sea surface in cross-polarization is only marginally reliant on incidence angles and sea surface wind speed. The dual-polarization (HH and HV) SAR data of RADARSAT-2, Sentinel-1, and Gaofen-3 constitute the basis for certain newly proposed glacier classifier algorithms [1, 4].

In this view, a cornerstone of both Antarctic and Arctic zonal deviations is the volume of sea ice prevalent. Sea ice exhibits reasonably notable spatiotemporal variability in the marginal ice zone (MIZ), which suggests that satellite measurements of ice cover at a fine spatial resolution than operational radiometer and scatterometer merchandise are crucial due to the speeding up declining trend of ice covers and diminished ice thickness in both Antarctic and Arctic zones [1–3, 5–7].

The question is now: what are the main algorithms exploited in retrieving ice covers in the SAR data? Despite the similarities between ice cover and the open ocean's SAR radar backscatter in certain circumstances, their textural elements that are dependent on brightness might fluctuate. To distinguish ice cover and deep ocean, descriptors are employed as well in conjunction with radar backscatter strength. The gray-level cooccurrence matrix (GLCM) texture characteristics can accurately capture distinctive backscatter (or brightness) features that are distinct across varying forms of the glacier and open sea, according to numerous research on texture examinations of SAR imagery [8, 9]. The textural characteristics of energy, contrast, correlation, homogeneity, entropy, and moment may be useful for classifying ice, according to earlier research [8–10].

Recent algorithms are exploited in glacier classifications based on the deep learning machine. Particularly, the support vector machine (SVM), a well-liked binary classification deep learning technique, has been focused on the identification of ocean glacier water (hence shortened to ice water) from spaceborne SAR data [1, 9, 11]. For numerous spaceborne SAR data, pixel-based and region-based glacier categorization methods have been invented [1, 4, 10, 12].

The critical question is: what are the disadvantages of using SVM in ice classifications in SAR data? Since the SVM employs a machine learning approach, acquiring a well-labeled training set that can distinguish between ice extent and deep ocean in Image data is probably the hardest process. Various glacier sorts, like multiyear glaciers, extent and distorted first-year ice, young ice, and numerous more, as well as various open ocean varieties, such as calm and rough sea surfaces, ought to be included in the training set. It takes a considerable amount of time and effort to acquire training datasets, which are often chosen extensively by professionals.

The novelty of this chapter is to form a novel algorithm based on the genetic algorithm (GA) for the automatic detection of ice covers in the Convair-580 aircraft. In actuality, the Convair-580 research aircraft serves as a multifunctional flying laboratory that supports a wide variety of studies. The normalized radar cross-section (NRC) Convair-580 is outfitted with cutting-edge technology for detecting aircraft physical parameters and atmospheric state (temperature, pressure, humidity, and three-dimensional wind).

#### **2. Convair-580 data acquisition**

Spectral ranges of 1000 nm to 2450 nm are covered by the 160-channel Short Wave Infrared (SWIR) hyperspectral imaging system. These data include a completely polarimetric dual-frequency (W and X-band) Doppler radar system. The Convair (**Figure 1**) will incorporate this NRC Airborne W and X-band radar system (NAWX) by January 2006. The thermal microwave emission from the surface and

*Introductory Chapter: Automatic Detection of Ice Covers in Airborne Radar Data Using… DOI: http://dx.doi.org/10.5772/intechopen.106982*

**Figure 1.** *Convair-580 research aircraft is used in this study.*

atmosphere is measured by the Airborne Multichannel Microwave Radiometer (AMMR) and is expressed in degrees Kelvin of brightness temperature. In the 1970s, the uplooking radiometer at 21 and 37 GHz, a part of AMMR, was created to monitor precipitation from an aircraft. The entire AMMR assembly operates between 10 and 92 GHz. Over the past three decades, a variety of aircraft has used the 21/37 GHz unit.

## **3. Marghany-based genetic algorithm (MBGA) for automatic detection of ice covers in Airborne SAR data**

This section introduces a developed genetic algorithm for ice cover automatic detection in any SAR data products. This algorithm is named as Marghany-Based Genetic Algorithm (MBGA) (**Figure 2**), which was adapted from the previous work of Marghany [15].

**Figure 2.** *Pseudo-code depicting MBGA algorithm.*

Let *K* be the total backscattered energy in the Convair-580 Aircraft data, and *β*<sup>1</sup> ð Þ, *β*<sup>2</sup> ð Þ, … , *β<sup>n</sup>* ½ � ð Þ be the sum of all backscattered bright pixels. Since genetic algorithms begin with the population initializing phase, *K* is composed of genes that represent the backscatter of bright pixels and their surroundings [13]. In this view, GA might be considered as having followed Sivanandam and Deepa [14] as:

Minimize

$$f(\boldsymbol{\beta}) = \begin{bmatrix} f\_1(\boldsymbol{\beta}), f\_2(\boldsymbol{\beta}), \dots, f\_k(\boldsymbol{\beta}) \end{bmatrix}^T \tag{1}$$

Formula (1) demonstrates that *fi* ð Þ *β* presents the *i-th* pixel backscatter *β* discrepancies in Convair-580 data, which signifies the *i-th* and *j-th* restraints backscatter energy in raw direction and column direction, respectively. Consequently, a fitness function is nominated to regulate the resemblance of separately backscatter energy associated with ice covers in the Convair-580 image. In this view, the backscatter of ice covers is signified by where *i* = 1,2,3, … , *K* and the initial population where *j* = 1,2,3, … , *N* and *i* = 1,2,3 … , *K*. Consequently, the fitness value of every separated population of the radar backscatter energy mathematical expressed as:

$$f(\mathcal{P}^{j}) = \left[\sum\_{i=1}^{N} \frac{\beta\_i}{N} \sum\_{i=1}^{K} \left| P\_i^j - \beta\_i \right| \right]^{-1} j = 1, \ldots, N. \tag{2}$$

In this perspective, *N* and *K* represent the number of the population involved in the fitness mechanism. In most scenarios, Convair-580 aircraft data exploits Eq. (2) to assess the level of similarity of phase corresponds associated with ice covers. Population sizes have been generated before this computation. Consequently, let us assume that *P<sup>j</sup> <sup>i</sup>* is a gene which corresponds to backscatter energy fluctuation through SAR data. Accordingly, *P<sup>j</sup> <sup>i</sup>* are chosen at random to illustrate the backscatter changes of the ice cover pixels as well as their surroundings. Additionally, *i* diverges from 1 to *K* and *j* fluctuates from 1 to *N* where *N* is the population dimensions.

In this understanding, let us consider the best fitness selection of individuals' backscatter energy *f P<sup>j</sup>* � � from the population *P<sup>j</sup> i* . Therefore, the maximum values of fitness of the population *Max f P<sup>j</sup>* � � and the minimum values of fitness of the population *Min f P<sup>j</sup>* � � are exploited to compute the threshold value *τ*, which is casted as:

$$
\pi = 0.5 \left[ \text{Max} f\left(\mathcal{P}^\circ\right) + \text{Min} f\left(\mathcal{P}^\circ\right) \right] \tag{3}
$$

The empirical formula (3) is employed as a phase in the selection process to establish the population's maximum and lowest fitness levels, respectively. In GA algorithms, this is regarded as the population generation phase for brightness patches in SAR data [15, 16].

The reproduction stage of a genetic algorithm, which incorporates crossover and mutation processes on the backscatter population *Pj <sup>i</sup>* in Convair-580 data, is primarily responsible for its operation. The crossover operator shapes the *P<sup>j</sup> <sup>i</sup>* to converge around solutions with high fitness. Therefore, the convergence occurs more rapidly the closer the crossover probability is to 1. The chromosomes exchange genes during the crossover process. Depending on the local fitness value, each gene can be formed by:

$$f\left(P\_i^j\right) = \left|\beta\_i - P\_i^j\right|\tag{4}$$

*Introductory Chapter: Automatic Detection of Ice Covers in Airborne Radar Data Using… DOI: http://dx.doi.org/10.5772/intechopen.106982*

In that circumstance, the crossover between two individuals process converts all individual populations of the first parent that have a local fitness *f Pj av* � � that is higher than the average local fitness and replaces the remaining genes with the matching ones from the second parent. Thus, the inclusion conditions characterize the average local fitness:

$$f\left(P\_{av}^{i}\right) = \frac{1}{K} \sum\_{i=1}^{K} \left| \beta\_i - P\_i^{i} \right| \tag{5}$$

The phenomenon of remarkable probability in the evolution process is thus denoted by the mutation operator. There is a potential that certain crucial genetic data about the chosen population will indeed be lost throughout the reproduction process. Hence, the mutation operator brings additional genetic sequences to the genetic diversity.

## **4. Results and discussion**

**Figure 3a** exhibitions a Convair-580 image with brightness patches designating the glacier zone. In contrast to the surroundings, **Figure 3b** reveals that the glacier patches seemed to have the maximum backscatter (�8 dB). The smallest backscatter, measuring �55 dB, is scattered among dark pixels that may represent calm water or a low wind zone.

The crossover process for the Convair-580 image is shown in **Figure 4** and involved 10 individuals. Positive bright spots in these 10 individuals represent the pixels that make up the ice cover, while negative dark patches represent the pixels that make up the surroundings, particularly still water or melting ice. Every cell is then compared to its counterpart in the other cells to determine if it is positive or negative. This study supports the work of Marghany [13] and Ninnis et al., [17].

**Figure 3.** *Data from a Convair-580 aircraft, shown as (a) composite bands of VV, HH, and HV and (b) backscatter variation.*

**Figure 4.** *First individual in the crossover phase.*

**Figure 5.** *Automatic detection of ice covers in SAR data using MBGA.*

**Figure 5**, consequently, demonstrates how the genetic algorithm can effectively isolate ice pixels from their surroundings. In other words, heavy ice covers, ice boundaries, and edges are all colored white, whereas calm water and melting ice pixels are all colored black (**Figure 2a**). This research is similar to Marghany's [13] previous work on using GA for object detection.

The significant question that arises is: what is the appropriate radar polarimetry for ice cover imaging? In this sense, **Figure 6** demonstrates that the HH band has lower standard errors of 17% than other bands VV and HV; respectively. On the contrary, HV has the highest standard error of 79%.

Consequently, the Marghany-Based Genetic Algorithm (MBGA) can also discriminate between different sorts of ice covers such as leads, young ice, and floes (**Figure 7**) in the HH band owing to its long tilt modulation as compared to VV and HV bands, respectively. In this sense, the receiver operating characteristic (ROC)

*Introductory Chapter: Automatic Detection of Ice Covers in Airborne Radar Data Using… DOI: http://dx.doi.org/10.5772/intechopen.106982*

**Figure 6.** *Standard errors among different polarized bands.*

**Figure 7.** *Automatic detection of different sorts of ice covers using MBGA.*

curve in **Figure 8** shows a significant difference in discrimination power between pixels representing leads, first-year ice, and floes with a probability *p* < 0.05.

The crossover procedure is what grants the MBGA its power. Each crossover process creates a new population. As a result, the fitness function scrutinizes multiple individual populations and incorporates them into subsequent populations. As a consequence, new populations are constantly generated based on the differences between two successive fitness values. Furthermore, the crossover procedure produces a more refined ice cover pattern by despeckling and preserving the morphology of the features of the ice cover groups through the fitness function used to implement the ice covers in different pixel classes. Indeed, the fitness function chooses a morphological pattern for the ice covers that is similar to the ice cover sorts that are recommended.

**Figure 8.** *ROC for ice sort discriminations using genetic algorithm (GA).*

## **5. Conclusion**

This chapter introduces the Marghany-Based Genetic Algorithm (MBGA), a novel algorithm for automatically detecting ice covers in SAR data. MBGA is thus designed as discrete steps of a modified genetic algorithm (GA). Individual backscatter fluctuation in SAR data is used as the primary source, which is generated by sequences of genetic algorithm procedures. In this scenario, MBGA outperforms other bands in terms of automatic detection of ice covers within the HH band, with the lowest standard errors of 17%. As a consequence, MBGA can automatically distinguish between different types of ice covers, such as leads, floes, and young ice. This is demonstrated using ROC, which indicates excellent discrimination of ice cover of different kinds with p < 0.05.

## **Author details**

Maged Marghany Global Geoinformation Sdn.Bhd., Masakiara Residences, Kuala Lumpur, Malaysia

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

© 2022 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.

*Introductory Chapter: Automatic Detection of Ice Covers in Airborne Radar Data Using… DOI: http://dx.doi.org/10.5772/intechopen.106982*

## **References**

[1] Li XM, Sun Y, Zhang Q. Extraction of sea ice cover by Sentinel-1 SAR based on support vector machine with unsupervised generation of training data. IEEE Transactions on Geoscience and Remote Sensing. 2020;**59**(4): 3040-3053

[2] Kwok R, Schweiger A, Rothrock DA, Pang S, Kottmeier C. Sea ice motion from satellite passive microwave imagery assessed with ERS SAR and buoy motions. Journal of Geophysical Research: Oceans. 1998;**103**(C4): 8191-8214

[3] Walker NP, Partington KC, Van Woert ML, Street TL. Arctic Sea ice type and concentration mapping using passive and active microwave sensors. IEEE Transactions on Geoscience and Remote Sensing. 2006;**44**(12): 3574-3584

[4] Minwei Z, Xiaoming L, Yongzheng R. The method study on automatic sea ice detection with GaoFen-3 synthetic aperture radar data in polar regions. 海洋学报. 2018;**40**(9):113-124

[5] Zakhvatkina N, Korosov A, Muckenhuber S, Sandven S, Babiker M. Operational algorithm for ice–water classification on dual-polarized RADARSAT-2 images. The Cryosphere. 2017;**11**(1):33-46

[6] Zakharov I, Power D, Howell M, Warren S. Improved detection of icebergs in sea ice with RADARSAT-2 polarimetric data. In: 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). Fort Worth, TX, USA: IEEE; 2017. pp. 2294-2297

[7] Silva TA, Bigg GR. Computer-based identification and tracking of Antarctic icebergs in SAR images. Remote Sensing of Environment. 2005;**94**(3): 287-297

[8] Clausi DA. Comparison and fusion of co-occurrence, Gabor and MRF texture features for classification of SAR Sea-ice imagery. Atmosphere-Ocean. 2001; **39**(3):183-194

[9] Clausi DA, Zhao Y. Grey level cooccurrence integrated algorithm (GLCIA): A superior computational method to rapidly determine cooccurrence probability texture features. Computers & Geosciences. 2003;**29**(7): 837-850

[10] Zakhvatkina NY, Alexandrov VY, Johannessen OM, Sandven S, Frolov IY. Classification of sea ice types in ENVISAT synthetic aperture radar images. IEEE Transactions on Geoscience and Remote Sensing. 2012; **51**(5):2587-2600

[11] Marghany M. Flock 1 data multiobjective evolutionary algorithm for turbulent flow detection. In: CD of 36th Asian Conf Remote Sensing: Fostering Resilient Growth in Asia. ACRS; 2015. pp. 1-6

[12] Tan W, Li J, Xu L, Chapman MA. Semiautomated segmentation of Sentinel-1 SAR imagery for mapping sea ice in Labrador coast. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2018;**11**(5):1419-1432

[13] Marghany M. Genetic algorithm for oil spill automatic detection from ENVISAT satellite data. In: International Conference on Computational Science and its Applications. Berlin, Heidelberg: Springer; 2013. pp. 587-598

[14] Sivanandam SN, Deepa SN. Introduction to Genetic Algorithms, by springer berlin heidelberg new york. 2008

[15] Marghany M. Utilization of a genetic algorithm for the automatic detection of oil spill from RADARSAT-2 SAR satellite data. Marine Pollution Bulletin. 2014;**89**(1-2):20-29

[16] Ressel R, Frost A, Lehner S. A neural network-based classification for sea ice types on X-band SAR images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2015;**8**(7):3672-3680

[17] Ninnis RM, Emery WJ, Collins MJ. Automated extraction of pack ice motion from advanced very high resolution radiometer imagery. Journal of Geophysical Research: Oceans. 1986; **91**(C9):10725-10734

## **Chapter 2**
