Based On Sub-aperture Polarimetric Sar Image Target Classification Algorithms

Polarimetric SAR(PolSAR) combines the high space resolution of SAR system and multiple channels data of the targets, which can reveal the physical scattering characteristics better. Therefore, extraction of characteristics and objects classification of PolSAR image play an important part in interpretation of radar data and target recognition. Using polarimetric information fully can express scattering mechanism much more exactly, which can obtain much better result of SAR image. In this dissertation, based on characters of PolSAR data and subaperture analysis, we research on target classification of PolSAR image. The main work and contributions accomplished in the dissertation are as follows:1. Firstly, the existing popular algorithms of PolSAR image classification and the problems are summarized. Besides, a new classification method, which employs AdaBoost algorithm, is proposed. Having known the types of objects in the scene, the classification accuracy is enhanced by using this method.2. The subaperture analysis of PolSAR image is studied. Firstly, based on time-frequency analysis, the decomposition theory of subaperture is investigated. Anisotropy polarimetric behavior and Bragg-resonance are also researched, and the latter is one of the important reasons for azimuthal polarimetric changes. Secondly, two nonstationary detection methods are studied. One is Maximum Likelihood ratio method, while the other combines entropy and the mean alpha angle, which is presented in this dissertation. By using this method, the nonstationary behavior of the azimuthal spectrum can be located and eliminated eventually.3. The target decomposition and the classification algorithms which are based on polarimetric decomposition are analyzed and simulated, such as Pauli decomposition,Krogager decomposition,H/α,H/A/α,H/α/Wishart,H/A/α/Wishart and Freeman-Wishart classification methods. What is more, a new algorithm, which combines the advantages of Pauli decomposition and AdaBoost algorithm, is proposed in this dissertation. This method can improve the effect of classification and speed up the convergence. 4. Three methods based on subaperture analysis and polarimetric decomposition are researched in this dissertation: the method based on the nonstationary detection and H/αplane, the method combined subaperture decomposition and H/α/Wishart classifier, and the method based on Freeman decomposition and subaperture scattering behavior.The results of simulating show that the image classification methods, which combine the subaperture data and polarimetric decomposition, can improve the effect of classification and speed up the convergence, and they are valuable tools for practical SAR image classification.