POL-SAR Image Classification Research And A Regression-Based Approach For Classifier Design

In this thesis, we do works on two kind issues of pattern reorganization, includingâ… . Apply classification method in POL-SAR image process;â…¡. Analysis the mechanism of classifier and propose a regression-based classier.In the first part, we focus on the research of Polarimetric scattering information representation and the terrain classification method for POL-SAR image, and do some works as follows.First, we do an analysis of the possibility of eigenvalue potential in Polarimetric information presentation. Specially, we use GMM model to simulate its distribution and use Bayesian classifier to demonstrate the feasibility and accuracy in Polarimetric information presentation.Second, we make an analysis of POL-SAR image features, both the nature property or the physical scattering features and its texture features, and point out the mutual complimentarily of these features. Some experiments are made to demonstrate the advantage of the combination features.Third, we provide a new eigenvalue-based approach for POL-SAR image classification. The method makes a selectively further SVM classification process on the basis of Bayesian classification, and achieves good results in POL-SAR image classification, which is a successful eigenvalue-based model for POL-SAR information presentation.The contributions of this thesis for POL-SAR image understanding include providing theory support for POL-SAR feature selection and exploiting the potential importance of eigenvalue in Polarimetric information understanding and representation.In the second part, we propose a regression-based classifier. With a deep analysis of classifier, we manage to find the restricted relationship between the features, which can be expressed by regression technical. Then we make some theory analysis and design a regression-based classifier. The key sense of this work is that it may build a transformation between classification and regression.This work was supported by the National Basic Research Program of China (No.61003198, No.60970067, No.60803097), and Central University Special Funds Basic Research and Operating Expenses (JY10000902038, JY10000902001, K50510020001å'ŒJY10000902045), Natural Science Foundation of Shaanxi Province (2009JQ8016).