The Research On Target Recognition Methods Based On Polarization Radar

With the progress of times and the development of science and technology, Radar whichis a kind of remote sensing tools is introduced into more and more use. Polarization syntheticaperture radar (SAR) as an important branch of radar development and the futuredevelopment direction, is one of most advanced sensors, has been widely concerned.Polarization SAR can describe target comprehensively, which has been applied to the targetrecognition field. Polarization SAR image classification and recognition is the core part ofSAR image interpretation, which has important application value in military and civildomains. At present, the polarization SAR image classification methods is still behind thedevelopment of polarization SAR system.In the system of radar target recognition, the target feature extraction and the design ofclassifier are the two key parts. This paper will focus on feature extraction of polarizationtarget and the better design of classifier.The work and the research results of this paper include as follows: In the second chapter,the basic polarization theory of the radar is introduced, including the electromagnetic wavepolarization equation and radar polarization equation. This is the theoretical basis of nextsteps. In the third chapter, we introduce the target feature extraction method which based onthe polarization decomposition. Target can be decomposed into several simple scatteringmechanisms base on the theory. Because there are many target polarization decompositionmethods, this chapter mainly introduces the Pauli decomposition, Krogager decompositionand Cloude-Pottier decomposition. Cloude-Pottier decomposition method which has beenforwarded by Cloude and Pottier is considered as the most successful polarizationdecomposition method by far. And through the experiments in this chapter, we verify theclassification recognition effect of polarization SAR image based on Cloude-Pottierdecomposition. In the fourth chapter, we introduce the SVM which based on structural riskminimization, and apply it to the research of polarization SAR image classification.According to the question of SVM classifier consuming huge time in the stage of training, wepropose the training sample optimal boundary vector algorithm based on K nearest neighbor(KNN) thought. This method can guarantee the final classification recognition effect. On the other hands, it can cut size of training samples, removes some redundant samples to shortenthe time of SVM training, achieve the purpose of optimizing SVM classifier finally. Theeffectiveness of the proposed method is verified by experiment. In the last chapter, wesummarize the work and discuss the direction of future research.