Classification Of PolSAR Image Based On Compressed Sensing

Polarimetric Synthetic Aperture Radar(PolSAR) is a kind of multi-parameter and multi-channel radar imaging system, it can describe the backscattering properties of targets by measuring the polarization scattering matrix under different polarization transceiver combinations, and can get more ground information than the traditional single-polarized aperture radar. Therefore, it has been widely used in the military, agriculture, water conservancy and other fields, and plays an important role.The image classification of PolSAR plays an important role in the interpretation of PolsSAR image, which is very important for the target detection and recognition of PolSAR.As feature extraction is the foundation of the classification of PolSAR image, this paper starts the research of the classification methods of PolSAR image from the aspect of feature extraction.First, the classical methods which named maximum likelihood classification and H/?/Wishart classification, and the methods which baesd on Support Vector Machine was analyzesd in the paper; then, some theories about compressed sensing was introduced; finally,the unsupervised method of classification based on Compressed Sensing and Support Vector Machine on this basis was proposed. In this method we first use the H/?/Wishart to select the traning samples, which in order to overcome the difficult problem during training samples in the classification methods; then we use the theory of compressed sensing to extract features from the original polarization coherency matrix; finally we use the Support Vector Machine to select the features and classify. Experimental results with data of Flevoland area from NASA/JPL AIRSAR system and Hayward area from UAVSAR system show that the novel method has a better classification results.