One-class classification for oil spill detection

SAR oil spill classification is a challenging topic, which is tackled by semi-empirical ad hoc approaches supported by very qualified experts. In all cases, the feature space is empirically defined, and two-class classification approaches are used. Although this approach allows achieving acceptable operational results, there is still room for improving both the comprehension of the physical phenomenon and the performance of classification techniques. In this paper, we propose a novel approach to oil-spill classification based on the paradigm of one-class classification. A classifier is trained using only examples of oil spills, instead of using oil spills and look-alikes, as in two-class approaches. Further, since the feature space is empirically defined, we also propose an objective technique to select the most powerful one that is suited for the oil-spill detection task at hand. Results on two case study datasets are reported to validate the proposed approach.