Research On Unsupervised Classification Method Of Polarimetric SAR Image

Polarimetric synthetic aperture radar(Polarimetric SAR) is multi-channel and active radar imaging system. itlaunchesdifferent combinations of polarimatric wave to ground, and then use an antenna to receive the target polarization scattering receive waves.Polarimetric SAR image processing extract the polarimetric information of target from the receiving scattering wave to reconstruct the observed scene.The polarimetric SAR image classification uses the scattering information of target to classify the target. This paper studies the unsupervised classification of polarimetic SAR image. In the absence of a priori information of the feature, only usingscattering information and statistical distribution information of obtained data for image classification, mainly studied the following three aspects:The Cloude-Pottier coherent decomposition based on second-order statistics polarimetric SAR data. Using the polarization scattering entropy H, average scattering angle?, anisotropy A classify the image. The Wishart classifier and the H-?-Wishart unsupervised classifier and the H-?-A-Wishart unsupervised classifier are analyzed. Considering the defects of the hard segmentation of the H- ? plane, the tolerable and intolerable areas are used to re-segement the H-? plane. For the tolerable area apply for the segementation rule of the H-? plane, and for the intolerable area apply for other segementation rule. This method avoids the defects and get a good result.The concept and properties of polarization similarity parameters are analyzed. Using the polarization similarity parameters to classify the polarization SAR image, then the advantages and disadvantages of the classifier are analyzed. Considering the disadvantage of polarization similarity parameters classifier, the polarization scattering entropy H was considered. In combination with the polarization similarity parameters and polarization scattering entropy H, a new classification method is considered and the results are improved.The polarization SAR coherent matrix decomposition theory based on scattering model is studied; especially the Freeman-Durden three-component decomposition and the disadvantages of the decomposition are studied.Theunsupervised terrain classification preserving polarimetric scattering characteristics based on the Freeman-Durden three-component decomposition and Wishart classifier is studied. Forthe shortcomings of Freeman-Durden three-component decomposition, a new decomposition method is proposed. This method combines the advantage of scattering model and coherent decomposition and solves the problem of Freeman-Durden three-component decomposition. Analysis showed thatclassification resultof this new method gets better classification than the Freeman-Durden decomposition methods.