Dark spot detection for characterization of marine surface slicks using UAVSAR quad-pol data

Abstract

Oil spills are a potential hazard, causing the deaths of millions of aquatic animals and this leaves a calamitous effect on the marine ecosystem. This research focuses on evaluating the potential of polarimetric parameters in discriminating the oil slick from water and also possible thicker/thinner zones within the slick. For this purpose, L-band UAVSAR quad-pol data of the Gulf of Mexico region is exploited. A total number of 19 polarimetric parameters are examined to study their behavior and ability in distinguishing oil slick from water and its own less or more oil accumulated zones. The simulation of compact-pol data from UAVSAR quad-pol data is carried out which has shown good performance in detection and discrimination of oil slick from water. To know the extent of separation between oil and water classes, a statistical separability analysis is carried out. The outcomes of each polarimetric parameter from separability analysis are then quantified with the radial basis function (RBF) supervised Support Vector Machine classifier followed with an accurate estimation of the results. Moreover, a comparison of the achieved and estimated accuracy has shown a significant drop in accuracy values. It has been observed that the highest accuracy is given by LHV compact-pol decomposition and coherency matrix with a classification accuracy of ~ 94.09% and ~ 94.60%, respectively. The proposed methodology has performed well in discriminating the oil slick by utilizing UAVSAR dataset for both quad-pol and compact-pol simulation.

Polarimetric Decomposition and Machine Learning‐Based Classification of L‐ and S‐Band Airborne SAR (LS‐ASAR) Data

“…Suppose a case of binary SVM (Chaudhary & Kumar, 2021), let the training data point be $$\left\{{x}_{1},{x}_{2},{x}_{3},{x}_{4}\text{\ldots }{x}_{n}\right\}$$. The decision function is given by Equation $$$F(x)=w.x+b$$$ $$w$$ → “weight vector”, which is orthogonal to the hyperplane, and $$b$$ → “bias” controls the hyperplane position at the maximum margin.…”

“…The most common kernel function used in SVM classification is linear, polynomial, radial bias function, and sigmoid. The RBF kernel is mostly preferable for classification (Chaudhary & Kumar, 2021; Mishra et al., 2017; Qian et al., 2014) due to its capacity to effectively handle the non‐linearity of training data. Equation represents the RBF kernel function.…”

“…Suppose a case of binary SVM (Chaudhary & Kumar, 2021), let the training data point be 𝐴𝐴 {𝑥𝑥1, 𝑥𝑥2, 𝑥𝑥3, 𝑥𝑥4. .…”

“…Researchers extensively utilize the machine learning algorithms such as random forest, decision tree, support vector machine, and neural for image classification due to their ability to handle multiple data sets and show better accuracy than other conventional classification methods (Luyi & Wen, 2016; Mishra et al., 2017; Yin et al., 2020). The support vector machine classifier is generally preferred for the classification (Chaudhary & Kumar, 2021; Lardeux et al., 2009; Niu & Ban, 2013; Sukawattanavijit et al., 2017) due to its least sensitivity to the dimensionality of the data sets (Pal & Foody, 2010), the classifier creates hyperplane boundary at the maximum margin between classes based on the input training sample data. The studies (Foody et al., 2006; Heydari & Mountrakis, 2018; Qian et al., 2014; Thanh Noi & Kappas, 2017) shows that the SVM classifier produces good accuracy in classification results even for small training sample data than other machine learning classifiers—artificial neural net, random forest, decision tree.…”

Cyber-physical oil spill monitoring and detection for offshore petroleum risk management service

“…The oil leaking accident is detected which can be analyzed using the remote sensing data. Synthetic Aperture Radar (SAR) is an effective and widely utilized remote sensing technique for oil spill detection 12 , 13 . During the detection procedure, an assumed oil leak source can be placed in a numerical simulation; and can be evaluated by comparing the simulation results with remote sensing data.…”