Polarimetric SAR Image Terrain Classification Based On Sparse Representation Classifier

POLSAR (Polarimetric Synthetic Aperture Radar) broaden the SAR (Synthetic Aperture Radar) applications. POLSAR image classification method has become an important part of remote sensing technology, which has become a hot research topic. Compared to conventional SAR image, POLSAR acquired HH, HV, VH, VV scattering echo information, which can get a more detailed feature information. With these new features added, we can improve the accuracy of classification of POLSAR.The main content of this thesis is to study the classification of POLSAR images. Classification is an important part of image interpretation technology, which plays an important role in cartography, glaciers recognition, distinguishing aspects of vegetation community. This thesis will focus on the polarization feature extraction, and then combining several classifiers for image classification, which mainly includes the following two aspects.1.Combined Stacked Autoencoder (SAE) and Support Vector Machine (SVM)) for image classification. This method can fully exploit the features of POLSAR information. Using a variety of methods to extract POLSAR scattering characteristics and polarization characteristics. Combine these two methods, that is, POLSAR eigenvalues. Place the extracted polarization scattering characteristics into stack autoencoder learning the more advanced features. Finally, use the SVM and Softmax for the final classification. Experiments demonstrate the effectiveness of the method.2.The proposed classification method based on polarization scattering characteristics and Sparse Representation Classification (SRC). Firstly, by Cloude decomposition, Freeman decomposition extracting coherent matrix, and finally entered into the SRC classifier, experimental results show that this method has higher classification accuracy, and has strong noise immunity.