Polarimetric Synthetic Aperture Radar Data Classification Using Sparse Representation

Polarimetric synthetic aperture radar (PolSAR) data contain a large amount of potential information that is very appropriate for terrain classification. In this paper we proposed sparse representation approach to classify PolSAR data. Among a large number of PolSAR parameters we have chosen the most optimum parameters to form feature vector. Using k-means algorithm feature space is divided to C clusters and their means form a dictionary. The performance of the proposed algorithm is compared with baseline methods such as SVM and neural networks. Experimental results show that the proposed method reaches good performance, especially for the regions where the baseline classifiers have weak performance. To assess the proposed algorithm, we employed Radarsat-2 fine-quad image acquired over San Francisco. KeywordsPolSAR image; classification; sparse representation, clustering