Feature selection and classification of polarimetric SAR images using SVM

A new feature selection algorithm is presented using SVM, and then it is integrated into the classification procedure of polarimetric SAR images to construct a novel SVM-based classification method. In the novel method, the sequential backward selection strategy is used to search feature subsets, and the number of support vectors is taken as the estimation index. Compared with those using the initial feature set and the classical RELIEF-F algorithm, higher classification accuracy with less or equivalent number of features is observed in a wider range of SVM parameters using the novel method.