Research on Multi-Feature Fusion for Discriminating Oil Spill and Look-Alike Spots

Oil spills and lookalikes (e.g. plant oil and oil emulsion) show the dark areas in SAR images, so it will bring some difficult in classifying the dark objects observed in full polarization SAR images into oil spills or lookalikes. In this paper, an approach is presented for distinguishing the dark areas in SAR images, which based on polarization features, geometric features and texture features fusion. In extracting the features, 18 different kinds of features consisting of 4 polarization features, 6 geometric features and 8 texture features are extracted. In the features analysis step, six features, which have good distinction, are selected to yield the optimal feature subsets. In the end, artificial neural network is used to classify the dark areas into oil spills or lookalikes. The proposed approach is applied to a data set of 14 oil spills and 6 lookalikes. The classification rate obtained by using the best set of features is 95%.