Terrain Classification Of PolSAR Image Based On SVM And Scattering Mechanism

Polarization synthetic aperture radar(pol SAR) has become one of the most important development direction in synthetic aperture radar at home and abroad. Compared with the single polarization radar image, Polarimetric synthetic aperture radar can provide more polarization information, So it can provide more information on interpreting and analyzing the radar images, That can make the interpretation of radar images quickly and accurately. Therefore, the study of pol SAR image classification has very important significance.In this paper, we mainly study the method of pol SAR image classification terrain based on SVM.The SVM based on statistical learning theory is an effective supervision classifier, it make use of the structural risk minimization principle, can obtain the optimal classification surfacein of minimal VC dimensional in a fixed risk at the same time,can avoid the overfitting problem,it has a good generalization performance, shows many unique advantages in solving the small sample, high dimension pattern recognition and nonlinear problems. At present, SVM has been applied in many fields, but research in full pol SAR image classification is still in its infancy, so this paper mainly studies the pol SAR image classification based on SVM, the main work of this paper can be summarized as follows:1.It put forward a method of terrain classification of pol SAR image based on SVM and Wishart measure. This method is mainly for polarization SAR coherent matrix and H/alpha features for classificiation, it’s mainly due to the polarization coherent matrix obeys, Wishart distribution from the angle of statistical distribution,and H/alpha n is an effective technology of polarization information extraction, the majority of experiments prove that the scattering entropy and the scattering angle in polarimetric SAR classification is very effective. So it will be combined the similarity measure based on coherent matrix and similarity measure based on polarization feature as the final SVM kernel function, the kernel function is not only using the basis of the statistical distribution knowledge of coherent matrix, but also joining the constraint information of target polarization scattering mechanism,can fully describe the similarity between target,thus can achieve better classification effect.2. It proposesd a method of terrain classification of pol SAR image based on k- means clustering and deep SVM. This method firstly uses k- means clustering to selete the effective information on the original training set as the final training sets to train the SVM classifier,this can greatly reduce the training set, and can effectively save the training and prediction time. then map the stacked SVM to multilayer, can get a deeper depth features and better classification accuracy,so can improve the accuracy of the pol SAR image and generalization.3. It proposesd a method of terrain classification of pol SAR image based on SVM of T matrix and texture feature.This method is mainly to study combining polarization features and texture features,and uses support vector machine(SVM) in pol SAR image classficiation. The polarization characteristics are mainly adopted in this paper is coherent matrix T of pol SAR image,it mainly due to the simple texture features for classification of pol SAR images without considering the polarization characteristics, So the edge of the classification effect is poorer,so combine two kinds of features effectively, Experiments show that this method is practical.