WEAKLY SUPERVISED POLARIMETRIC SAR IMAGE CLASSIFICATION WITH MULTI-MODAL MARKOV ASPECT MODEL

In this paper, we present a weakly supervised classification method for a large polarimetric SAR (PolSAR) imagery using multi-modal markov aspect model (MMAM). Given a training set of subimages with the corresponding semantic concepts defined by the user, learning is based on markov aspect model which captures spatial coherence and thematic coherence. Classification experiments on RadarSat-2 PolSAR data of Flevoland in Netherlands show that this approach improves region discrimination and produces satisfactory results. Furthermore, multiple diverse features can be efficiently combined with multi-modal aspect model to further improve the classification accuracy.