Detection and classification of oil spill and look-alike spots from SAR imagery using an Artificial Neural Network

Oil spills represent a major threat to ocean ecosystems and their health. The recent incident in the Gulf of Mexico demonstrates the potentially catastrophic nature of offshore oil spills. Illicit pollution requires continuous monitoring and satellite remote sensing technology represents an attractive option for operational oil spill detection. Previous studies have shown that active microwave satellite sensors, particularly Synthetic Aperture Radar (SAR) can be effectively used for the detection and classification of oil spills. Oil spills appear as dark spots in SAR images. However, similar dark spots may arise from a range of unrelated meteorological and oceanographic phenomena, resulting in misidentification. A major focus of research in this area is the development of algorithms to distinguish oil spills from `look-alikes'. This paper describes the development of a new approach to SAR oil spill detection using two different Artificial Neural Networks (ANN). The first ANN segments a SAR image to identify pixels belonging to candidate oil spill features. A set of statistical feature parameters are then extracted and divided into subsets to facilitate sensitivity analyses. The second ANN classifies objects into oil spills or look-alikes according to their feature parameters. A pilot study employed sixty-two ERS-2 SAR and ENVSAT ASAR images of verified oil spills or look-alikes to train and evaluate the algorithm. Overall accuracies of 96.52 % were obtained for pixel segmentation and 95.2 % for feature classification. The segmentation approach outperformed established edge detection and adaptive thresholding techniques. An analysis of feature descriptors in the classification stage highlighted the importance of image gradient information.