Unsupervised classification of polarimetric SAR Image by Quad-tree Segment and SVM

This paper presents a new method for unsupervised terrain classification using fully polarimetric synthetic aperture radar image based on quad-tree segment and support vector machine techniques. This unsupervised classification method begins with quad-tree segment technique that ensures each segment contains the data of only one cluster. Then, the feature vectors are constructed by sampling those segments using polarimetric covariance matrix. Then, the feature vectors of the samples of every segment are selected to create training sets based on the number of support vector, which is obtained by support vector machine technique. Finally, the fully polarimetric SAR data is classified using the support vector machine trained by those training sets, and the classification results are analyzed in detail. According to the classification results, it is shown that the method presented in this paper have sound classification accuracy and flexibility.