Trained and nontrained CFAR detection of oil slicks on the ocean surface by resorting to SAR data

This paper addresses detection of oil slicks on the sea surface based on possibly multifrequency SAR data. Two detection strategies have been derived by resorting to the GLRT: the former relies on a set of training data, namely returns from pixels corresponding to a slick-free area; the latter, instead, does not assume the availability of training data. Both strategies can be implemented by resorting to either single or multifrequency data and all of the proposed implementations guarantee the CFAR property. The performance assessment, based on SIR-C/X-SAR data, shows that non-trained algorithms can guarantee satisfactory performance in cases of relevant practical interest.