EFFICIENT LAND COVER CLASSIFICATION TECHNIQUES OF FULLY-POLARIMETRIC SAR IMAGES

The present work proposes three methods for the classification of the polarimetric synthetic aperture radar (PolSAR) images; namely the methods of likelihood ratio test (LRT), covariance matrix matching with continuously sliding window (CMMC) and covariance matrix matching with jumping sliding window (CMMJ). The three methods lie within the category of supervised classification techniques. For performance assessment, each of the three methods is applied to two sample simulated quad-channel images similar to those obtained from real PolSAR systems. The performance of the classification methods are assessed using the following performance measures: the confusion matrix (CM), producer's accuracy (PA), user's accuracy (UA), overall accuracy (OA), false alarm ratio (FAR), and kappa coefficient (K). Also, the present work proposes a false alarm correction (FAC) method to correct the classification image resulting from the application of each of the proposed classification methods. The dependence of the performance of the proposed classification methods on the resolution of the quad-channel PolSAR image is investigated. It is shown that the application of any of the proposed classification techniques followed by the FAC method significantly improves the performance of the overall classification process and results in accurate classification of the land covers included in the PolSAR image.