An SVM-based method for land and sea segmentation in polarimetric SAR images

A Support Vector Machine (SVM) based method for land and sea segmentation in Polarimetric SAR (POLSAR) is proposed in this paper. The principle of SVM is first briefly summarized. Features that selected for SVM consist of 9 polarimetric features obtained from polarimetric target decompositions, i.e., Krogager, Freeman-Durden and Cloude decompositions, and 6 texture features calculated from first-order statistics. These 15 features are combined to feature vectors. The experiments are carried out on POLSAR data from Radarsat-2. The SVM classifier is obtained through training with selected land and sea samples and then applied in segmentation of the images to be tested. The segmentation results indicate the effectiveness of the proposed method. The results are analyzed and the parameter selection of SVM is discussed in brief.