Oil spill detection using synthetic aperture radar images and feature selection in shape space

The major goal of the present study is to describe a method by which synthetic aperture radar (SAR) images of oil spills can be discriminated from other phenomena of similar appearance. The optimal features of these dark formations are here identified. Because different materials have different physical properties, they form different shapes. In this case, oil films and lookalike materials have different fluid properties. In this paper, 9 shape features with a total of 95 eigenvalues were selected. Using differential evolution feature selection (DEFS), similar eigenvalues were extracted from total space of oil spills and lookalike phenomena. This process assumes that these similar eigenvalues impair classification. These similar eigenvalues are removed from the total space, and the important eigenvalues (IEs), those useful to the discrimination of the targets, are identified. At least 30 eigenvalues were found to be inappropriate for classification of our shape spaces. The proposed method was found to be capable of facilitating the selection of the top 50 IEs. This allows more accurate classification. Here, accuracy reached 94%. The results of the experiment show that this novel method performs well. It could also be made available to teams across the world very easily.