**5. Existing methodologies**

A review of literature survey on various methodologies used to detect oil spill for ocean monitoring has been carried out and presented subsequently in detail.

In 1999, Solberg et al. [31] have proposed an algorithm for automatic detection of oil spill using rule-based approach combined with statistical modeling (i.e., Gaussian density). It also uses prior knowledge about oil slick for classification of it. In this algorithm, authors have used 11 features of oil slick that include distance to point source, number of detected spots in the scene, number of neighboring spots, Homogeneity, Slick Complexity, Slick width, slick area, first invariant planar, local area contrast, border gradient, power-to-mean ratio. The accuracy attained by rulebased classification approach is 95%. But the observed limitation of this approach is it requires prior knowledge about oil slicks. The proposed algorithm could be considered as semi-automatic algorithm as it uses prior knowledge.

In 2000, Fabio Del Frate et al. have proposed a neural network algorithm for classification of oil spill from look-alike in [1]. The proposed algorithm uses the rule-based Bayesian statistical decision approach. The performance of classification algorithm has been evaluated on a dataset containing oil spill and look-alike. The proposed algorithm uses some of the features that describe both physical and geometrical properties of oil spill. Those features are used as candidate features for classification. The accuracy attained by proposed algorithm is 70%. However, the observed limitation of this approach is that it is required to develop the classification rules for the complex datasets. Hence, this algorithm could be considered as neural network algorithm as it uses statistical-based classification decisions.

In 2004, Maged Marghany [32] has proposed an approach for oil slick detection and oil slick trajectory model. The proposed approach uses two sub-models. The first model uses texture analysis for oil slick detection. The second model uses the oil slick trajectory forecasting model. The oil slick trajectory model contains the integration between Doppler frequency shift model and Lagrangian model. Both models have used a classical Fay's algorithm, to simulate the oil slick trajectory movement. In this proposed algorithm, authors have used features such as entropy, surface current changes, energy, slick homogeneity, and slick direction. The results of this approach are not accurate, which is a primary limitation. The proposed approach could be considered as an automatic algorithm as it uses slick trajectory models of classical Fay's algorithm.

In 2004, Frederic Galland et al. [33] have proposed an approach for monitoring pollutant of major environmental hazards at ocean surface and minimizing it. The proposed approach consists of two parts. In first part of the proposed algorithm, homogeneous regions are partitioned that should be applied with polygonal active grid. In second part of the proposed algorithm, classification is carried out and it should be applied with an automatic minimum description length (MDL) thresholding technique. In this work, feature considered includes size, shape, distance of object from coastal region, intensity, and texture. The results of accuracy of oil spill regions are obtained

for larger areas with appropriate coastlines. In this proposed method, observed limitation of MDL application is that noise is not removed fully from SAR images.

In 2005, Iphigenia et al. [34] have proposed computational intelligence-based approach for oil spill detection on satellite images. In this approach, smoothing is performed through Gaussian filtering followed by thresholding is applied. The image enhancement technique is applied to enhance dark regions and then, objects are grouped together using segmentation method. Finally, fuzzy classification approach is applied to classify oil spill or look-alike. In this approach, authors have used three features: region eccentricity, land distance and the probability of oil spill. The proposed algorithm has been tested with 26 images and attained 88% of accuracy in detecting oil spill regions.

In 2005, Fanny Girard-Ardhuin et al. [35] have proposed semi-automatic approach for detecting spilled oil regions, characterization of region properties, and classification of oil spill or look-alike. For semi-automatic detection of dark regions median filtering is applied to remove noise followed by Sobel operator for thresholding and segmentation. The filtered regions are characterized using features such as size and shape. In this algorithm authors have used some of the features about the oil spill formation of wavelength, spill polarization of sea surface areas, satellite projection of incidence angles with radar projection view, the slick nature and changes of the sea surface conditions. While applying this algorithm, the image dark region parameters are not properly detected and its complexity is not determined accurately.

In 2006, Gregoire Mercier et al. [36] have proposed a semi-supervised approach for oil spill detection using wavelet decomposition. This proposed method is implemented in complex sea surface and features used are sea surface wavelength, spill polarization of image region, the radar different views of oil spill projection. The obtained results of oil spill are shown as very effective and accurate in detecting oil slicks from the specified image region. This proposed approach is applicable for large ENVISAT images with complex formatting. It is considered as limitation of this algorithm.

In 2007, Topouzlis et al. [37] have proposed a neural network-based approach for detecting and classifying oil spill. The proposed approach detects the dark formation areas and classifies it into oil spill or look-alike. In this proposed algorithm, features such as oil spill perimeter, shape factor of the region, the object complexity estimation, standard deviation, and power-to-mean ratio are used. The obtained results show that the algorithm detects dark formation with 94% accuracy and discriminate it as oil spill with 89% accuracy. The limitation of this proposed algorithm is that it is not suitable for high dynamic data environments.

In 2007, Solberg et al. [38] and Maurizio Migliaccio et al. [39] have proposed a three-step algorithm for detection of oil spill using the RADARSAT and ENVISAT SAR images. The first step is used to detect the dark spot and the second step is used to extract the dark spot features, and finally, it classifies the dark spot as oil spill or look-alike. In this proposed algorithm, authors used features such as the complexity, width, area, and slick moment; contrast of dark region about local contrast, border gradient and smoothness of the spill region; homogeneity of spill region about the power-to-mean ratio; the spill surroundings about the detection of dark spots, spot region, and distance from the coastline. The result of this algorithm is reported for 59 images of RADARSAT and ENVISAT about oil spill or look-alike. The observed limitation is classification approach not carried automatically.

In 2008, Lena Chang et al. [40] have proposed an approach for ocean monitoring with minimized computational complexity. In this proposal, image segmentation is carried out using split and merge technique. The conventional detection theory called

### *Computational Techniques of Oil Spill Detection in Synthetic Aperture Radar Data: Review Cases DOI: http://dx.doi.org/10.5772/intechopen.108115*

centralized look-alikes ratio test (GLRT) is used to detect oil spill using simple variance and statistical properties with the help of decision-making rules. The proposed algorithm determines oil spill effectively on ERS-2 SAR images. The similar pixels are not shown on the specified image region, while applying the GLRT method using ERS-2 SAR images.

In 2008, Camilla Brekke et al. [41] have proposed improved classification algorithm using a multivariate Gaussian classifier for reducing low-confidence levels. In this work, authors have used features such as slick complexity, power-to-mean ratio of slick, slick border, slick local contrast, slick width, slick region, slick smoothing contrast estimation, and slick variance. The observed limitation of this algorithm is that the manual operation could not be performed on selected region. In Ivanov et al.'s work [42] image processing methods are proposed for detecting oil spill. In this proposal, study is carried out on East China Sea for oil spill detection using SAR images. Authors have selected features such as shape, slick size or length, location, orientation, type of edge on oil slick, and database contrast. The result and analysis show small oil patches are uniformly distributed along the ship tracking. From the results, detection of oil spills on image region is not accurate. Kontantinos N. Topouzelis [43] have proposed a classification with fuzzy logic technique. In this proposal, authors have used nearly 25 features of oil spill includes object's area, perimeter, perimeterto-area ratio, complexity, shape factor I, shape factor II, mean value, standard deviation, power-to-mean ratio, background mean value, background standard deviation, background power-to-mean ratio, ratio of the power to mean ratios, mean contrast, max contrast, mean contrast ratio, standard deviation contrast ratio, local area contrast ratio, mean border gradient, standard deviation border gradient, max border gradient, mean difference to neighbors, spectral texture, shape texture, and mean Horlicks texture. Authors have claimed that the obtained accuracy is reported as 99% in detection of oil spill. But this algorithm observed some limitations that are the imbalanced training datasets, data selection validity, and high dynamic environment selection of dataset. In Chuanmin, Hu et al.'s [44] works use slick region statistical parameters like mean, standard deviation, and minimum and maximum distance of the slick from the region. The results show an annual seapage rates in the northwestern Gulf of Mexico and also the slick time series variation recording. But the observed limitation is that it could be performed only on medium resolution of MODIS data under tropical and sub-tropical.

In 2010, Konstantinos Topouzelis et al. [45] have proposed an algorithm for feature selection and classification. In this algorithm, classification method uses the statistical decision tree rules and features considered are area, perimeter, complexity, shape, mean, power-to-mean ratio, texture, and contrast of oil spill region. The result shows that the obtained classification accuracy rate is 84.4% in detecting oil spill. This proposed method suffers from problem of classification of 70 trees and it performs only sequential selection operation.

In 2011, Biao Zhang et al. [46] have used an unsupervised classification approach for distinguishing oil slicks from sea surfaces. It uses very simple and effective mapping techniques. In this proposal, quad polarization SAR images of lower wind condition are mapped with normal condition by considering features such as slick entropy and slick region. But the observed limitation is the selected smaller dark regions on images could be processed only under moderate wind conditions on sea surface regions.

In 2012, Chanudhuri et al. [47] have proposed four step algorithms for automatic detection of ocean disturbance. This proposed approach first enhances SAR images to emphasize the dark regions. The second step then segments dark regions using

iterative method. After this, the graph theory-based method uses to remove the unwanted false alarms. Finally, a link-based algorithm is applied for detecting the disturbance features using statistical approaches. In this algorithm, the used features enhance the image, segmentation of the image regions, hole filling, region removal of slick enhancement, segment link of the same regions, and spur removal of specified region. The result shows that synthetic test images with different noise levels and real images from a variety of satellites such as ERS-2, SEASAT, ENVISAT, RADARSAT for sea disturbance features. But this algorithm observed one limitation to compute the data models in a single way only. Bhogle et al. [48] have proposed algorithm for automatic detection of oil spills. This algorithm consists of three steps for protecting the water life and reduces environmental damages from oil spills. In this algorithm, the first step detects the dark spot on image regions. The next step is feature's extraction of each dark spot. The final step is classification of dark spot is oil spill or lookalike. After these steps, then the texture entropy is used for discrimination of oil spill area and Mahalanobis classifier is estimation of the oil spill identification. In this algorithm, authors used three features of oil spill such as the distance from coastline, mean of spill on specified region, and the co-variance of image region pixels. The reported result shows different algorithm for an automatic detection with accurate and efficient manner. And also computing the texture entropy features 94% of accuracy with help of Mahalanobis classifier. But this algorithm observed some limitations are texture entropy might complex process and less flexible using the Mahalanobis classifier is not effectively remove the speckle noise for data analysis. This proposed algorithm classifies the texture variations of spill detection on image regions.

In 2012, Michele Vespe et al. [49] have proposed approach for quality issues of satellite images related to marine applications. This algorithm is to assess the quality and indicators for vessel and oil spill detection. This algorithm used some features including global quality aspects of radiometric sensitivity, radiometric resolution, radiometric accuracy, region stability and error to detecting, spatial resolution, geolocation accuracy about the local-quality aspects with peak-to-sidelobe ratio, the integrated sidelobe ratio, ambiguity of slick detection, interference of artifacts, structured data, missing data, and SAR processing errors. The result is reported based on current issues of marine applications from oil spill detections. But the observed algorithm limitation to generalize the quality levels should be encountered based on the data representation.

In 2014, Alireza Taravat et al. [50] have proposed neural networks algorithm for fully automatic detection of darkspots. In this algorithm, two methods use nonadaptive identification of oil spill on image region. First method called Weibull Multiplicative Model (WMM) is applied for filtering in each sub-image. Second method called Pulse-Coupled Neural Networks (PCNN) is used for segment and removing the false targets. In this algorithm, authors used features as well-defined slick region identified, linear well-defined, massive well-defined, not well-defined, linear not well-defined, massive not well-defined, linear dark spot, and massive dark spot. The results tested 60 ENVISAT and ERS-2 images and reported 93.66% accuracy of overall dataset. But algorithm-observed limitation shows a fewer effective result.

In 2014, Yu Li et al. [51] have proposed a statistical quad-pol reconstruction method for SAR quad-polarization data. This method improves a compact polarimetric SAR data for reconstruction using statistical methods. This proposed method refines the scatter behavior of SAR signal relationship for iterative quad-pol reconstruction data. In this algorithm, authors used only one feature that is slick region iteration and the reconstruction of quad-pol data. The result shown is to improve

### *Computational Techniques of Oil Spill Detection in Synthetic Aperture Radar Data: Review Cases DOI: http://dx.doi.org/10.5772/intechopen.108115*

compact polarimetric SAR data using the quad-pol reconstruction model accurately. But the algorithm observed one limitation is to perform only for low-accuracy quadpol reconstruction model. This proposed algorithm has been used to improve the data performance and data accessing model for reconstruction using SAR polarimetric techniques.

In 2014, Salberg et al. [5] have proposed algorithm that uses hybrid polarimetric SAR technique for slick classification. This proposed algorithm is used only to improve hybrid-polarimetric SAR data and its feature extraction. In this algorithm, authors used five features that include oil spill detection on image dark regions, correlation co-efficiency, standard deviation, slick decomposition of region, conformity index of the slick with polarimetric SAR data, and coherence measurement of slick region separation. The result shows the features of hybrid polarimetric SAR data and then classifies each detected slick on the specified region. But algorithm observed one limitation that performs only limited testing images from the hybrid polarimetric SAR data. This proposed technique identifies slick feature and its measuring capabilities of hybrid polarimetric SAR data.

In 2014, Giacomo De Carolis et al. [52] have proposed an approach for measuring the slick thickness in medium-resolution imaging spectrometer instrument (MERIS) and moderate-resolution imaging spectro-radiometer (MODIS) images. This proposed approach estimates the thickness of oil slick from marine surface. In this proposed approach, authors used one feature for estimation of slick thickness from the detected region. The result shown is reported from the June to August 2006 under Lebanon oil spills occurred in MERIS and MODIS gathered data. But this approach not shows the perfect values of gathered data samples as a limitation.

In 2015, Collins et al. [53] have proposed approach as a compact polarimetric SAR technique for coherent dual-pol SAR images. This approach uses simulated airborne compact SAR data for characterizing oil-water mixing of deep-water horizon oil spill. On the other hand, this SAR is used to perform the great potential for maritime surveillance application on oil spill characterization. In this approach, author used single feature as reconstructing the errors for sea water. This result shows quad-pol data features and pseudo-quad data and their differences. But this approach is observed only for variations of different compact polarimetric SAR images.

In 2016, Saeed Chehresa et al. [54] have proposed an algorithm to detect the oil spills and lookalikes based on the selection of optimum features in a given SAR data set('s). This algorithm is evaluated based on classification of SAR image dark spots. This algorithm of 93.19% accurate is classified with optimum set of features from the dataset. Also, eight different evolutionary algorithms are considered to classify the desired feature subsets. Giacomo Capizzi et al. [55] have proposed an automate clusterbased system developed for oil spill detection in satellite remote sensing. This system interactively working with several characteristics about the availability and adaptability of the different classes of objects and SAR images are considered based on application areas. This proposed algorithm uses a back-propagation neural network algorithm to obtain the outcomes as identify objects such as ships with spills/sliks on SAR images.

In 2017, M. Konik and K. Bradtke [56] have proposed an object-oriented approach to detect oil spills on ENVISAT ASAR (Advanced Synthetic Aperture Radar) images. This proposed methodology improves the classification at the scale of entire water bodies, focusing on its repeatability. Also, this approach analysis enhances the optimized filters to multilevel hierarchical segmentation. This proposed system recorded 96.15% of accurately identified spills and 4% of dark spots extracted from the given dataset.

In 2018, Majidi Nezhad et al. [57] carried case study on sentinel 2 satellite images for oil spill detection analysis. In this analysis part, exploit areas of oil spills and oil loading ports are often used by the governance of tankers and ship. With ENVI tool, this analysis is carried out to discuss oil spill detection in Persian Gulf by using multisensor images data.

In 2019, Jiao et al. [58] proposed a deep learning approach to detect the oil spills in an unmanned aerial vehicle. This proposed approach having three main steps, first step used with CNN model to detect oil spills in image. Second step is used for filtering the detected results of first step with help of Otsu algorithm. Third step is used to detect detail manner in a region-based detection with oil spills. So, this proposed approach is effectively solved to reduce the cost of oil spill detection by 57.2% compared with the traditional inspection process.

In 2020, Yekeen et al. [59] proposed a novel deep learning algorithm to detect the oil spills. This process can be achieved by the similar visuals of oil slicks and look-alike, which affects the reliable SAR images for the marine oil spill detection. But, this can be used for minimal detection and discrimination of oil spills with help of traditional deep learning models with limited accuracy. Moreover, the proposed novel deep learning algorithm detects the oil spills using computer vision instance segmentation mask region based on CNN model to detect maximum oil spill on ocean regions.

In 2021, Syedi et al. [60] have proposed multiscale multidimensional residual kernel convolution neural network. This proposed method is used for overall accuracy of oil spill detection in Gulf of Mexico regions. Also, this study investigates a proposed model with OSD algorithms had shown better performance.
