Object-oriented Polarimetric Decomposition And Feature Description Of High Resolution SAR Image

With the development of high resolution Polarimetric Synthetic ApertureRadar (PolSAR) images, the processing methods of PolSAR images haveevolved from pixel-oriented analysis to object-oriented analysis. The mainpurpose of object-oriented analysis is extracting various textural and spatialfeatures from image objects to improve classification accuracy. Thepolarimetric features of the existing object-oriented analysis are extractedbased on pixel instead of object, which leads to large computation and lowefficiency. To solve this problem, a object-based polarimetric decompositionmethod is proposed. The coherent matrix of object is constructed by weightediteration of scattering coefficient of similarity, and the convergence ofcoherent matrix is analyzed, polarimetric features could be obtained throughthe decomposition of object’s coherent matrix. Comparisons withthe pixel-oriented classification method and object-oriented classificationmethod verify that the proposed method can improve the efficiency ofpolarimetric features extraction and increase the accuracy of terrain targetsinterpretation. For several typical terrain targets, a decision tree algorithm isutilized for evaluating the contribution of different features, which usesRADARSAT-2PolSAR data. Further investigation is carried out on the respective description of the spatial features and the polarimetric features,which provides important reference for establishing typical terrain targetsinterpretation database of high resolution SAR.