Classification of Oil Spills and Look-alikes from SAR Images Using Artificial Neural Network

Abstract

An active sensor imaging technology like Synthetic Aperture Radar (SAR) due to its all-day, all-weather imaging capability, can be effectively used for the detection of oil spills over the ocean. Oil spills, due to their low backscattered energy towards the SAR sensor, appear as dark spots/areas in SAR images. However, some other meteorological and oceanographic phenomena also have same spectral signature and appear as dark spots resulting in misidentification. Hence, a major challenge in oil spill detection using SAR images is to discriminate between oil spills and look-alikes. The success of correct oil spill detection depends on the feature extraction of all the dark spots and accurately distinguishing as oil spills and look-alikes based on their features. This paper describes the development of an Artificial Neural Network for classification of oil spills and look-alikes using some geometric and radiometric features vector. A Sentinel-1 image dataset having 51 images with 31 oil spills and 20 look-alike scenes is used to train and evaluate the classifier performance. Overall accuracy in the range 90 % - 95 % was obtained for classification.