Research On POLSAR Images Classification Based On Target Decomposition

Polarimetric synthetic aperture radar (referred to as POLSAR) is widely used in land cover classification. As the basis for POLSAR image classification, there exist polarization, texture, color and other aspects characteristics in the POLSAR images. The different characteristics is the embodiment of target different aspects of nature, and they can provide different information on the classification processing. Integrating polarization, texture, color features for classification can improve the classification accuracy of POLSAR images. Based on this purpose, this paper analyzes the extraction method of polarization, texture, color features, and make the combination of the texture and color features with polarization feature, and the main work as following:(1) Based on a variety of polarimetric target decomposition method,it obtains many polarization parameters, then compose 6 different polarization feature vectors, which is used for POLSAR image classification, and then make analysis of the effect of different polarization characteristics parameters on the classification results.(2) Based on gray image,it gets 6 texture feature parameters then construct texture feature vector. The texture feature vector and the 6 polarization feature vectors were combined, forming 6 new feature vectors. Classification results show that join with texture features can improve the classification accuracy of POLSAR image.(3) Based on the pseudo color enhancement image of gray image,it extracts color histogram parameters to construct color feature vectors. The color feature vector and the 6 polarization feature vectors were combined, forming 6 new feature vectors. Results show that combining with color feature can improve classification accuracy of POLSAR image.(4) Combining the characteristics of polarization, texture, and color features to form 6 vectors for POLSAR image classification, and have proven to be superior. Classification results shows that the characteristics of polarization, texture, color classification of POLSAR image is a processing of mutual to promote the overall effect, the join of texture features and color features can improve classification accuracy.