Oil Spill Characterization and Monitoring Using SYMLET Analysis from Synthetic-Aperture Radar Images

Background: To identify and monitor the occurrence of oil spill in the ocean using image data obtained from different satellites. It is to characterize and identify type and other physical characteristics of oil using appropriate image processing techniques and to predict the physical characteristics of oil spill using machine learning methods. Methods: In this paper, forty sample images were used with wavelet transform analysis and machine learning techniques. For wavelet analysis (SYMLET family such as sym2, sym3, sym4, sym5, sym6, sym7, sym8 analysis) and machine learning, k-nearest neighbor algorithm is applied to optimize the oil spill feature sets. Features included RGB, spreading, complexity, standard deviation, entropy, ellipticity, intensity and correlation coefficient. This experiment was conducted on RADARSAT-2 SAR images. The features were classified using k-nearest neighbor algorithm. Seventy percent of features used for training and thirty percent for testing. Results: The results show that oil spill classification achieved by wavelet transforms and machine learning algorithms outperformed very well with similar parameter settings, especially with 70% training data and 30% testing data using confusion matrix. It also represents 92.6581% accuracy for crude oil using SYMLET 5 analysis which indicates better characterization of oil spills. Results denote oil spill detection using synthetic-aperture radar (SAR) remote sensing which provides an excellent tool in oil spill characterization; various features can be extracted from SAR data set.