Research On Polarimetric Target Decomposition Of PolSAR Image And Its Application

Polarimetric SAR （PolSAR） image contains lots of useful informationrelated with the physical, chemical and geometric properties of the targets. Themodel-based non-coherent target decomposition theorem is one of the usefultools to extract target information from PolSAR image. The algorithm of thistheorem is simple and the results are easy to interpret and have clear physicalmeanings. Based on the analysis of the problems of the current non-coherenttarget decomposition methods, this paper proposed the improved targetdecomposition methods.Focusing on the problems such as restriction of the symmetric reflectionassumption, polarimetric information is not fully utilized and the roughcharacteristics of the man-made targets, this paper first introduced the basictheory of the PolSAR technique and several classic target decompositionmethods.Secondly, the asymmetric scattering term is introduced by analyzing thelimitation of the helix scattering in representing the asymmetric information ofthe man-made targets. Based on asymmetric scattering term, asymmetricdecomposition method based on orientation-angle-compensated coherencymatrix and asymmetric decomposition based on covariance matrix is proposed.These two methods successfully solved the limitation of the symmetric reflectionassumption and fully utilized all the polarimetric information. To solve theproblem of the rough characteristics of the man-made targets, two-layerdecomposition method is proposed to extract the detail scattering information ofthe man-made targets. It has been found that the ratio of the asymmetricscattering power to the total power equals to the normalized correlationcoefficient between co-polar and cross-polar channels. This paper also proposedthe terrain classification method based on Support Vector Machines （SVM） andtwo-layer decomposition to research the application of target decompositiontheorem in terrain classification.Then, the orientation angle compensation method has been extended and four other angle decomposition methods based on coherency matrix have beenproposed, which are ellipticity angle compensation, quasi-ellipticity anglecompensation, absolute phase compensation and quasi-absolute phasecompensation. These four proposed angel compensation methods caneliminateIm[T₁₃],Re[T₁₃],Im[T₁₂]andRe[T₁₂], respectively. With thesupplement of the proposed four decomposition methods, there exist six anglecompensation methods that can be selected to eliminate the corresponding sixnon-diagonal elements of the coherency matrix.Finally, the experimental analysis has been taken to evaluate the proposedmethods using L-band PolSAR data. Experimental results indicate that both theasymmetric decomposition method based on orientation-angle-compensatedcoherency matrix and the asymmetric decomposition based on covariance matrixcan be effectively used to extract the information of the man-made targets. Thesetwo decomposition methods can be successfully applied in both naturaldistributed areas and urban areas. What’s more, the results of the two-layerdecomposition show that different buildings have different asymmetric scatteringcomponents, which have significant potentials in the study of classification andrecognition of the man-made targets. The experimental terrain has beeneffectively classified by the classification based on SVM and two-layerdecomposition method. Besides, the results of the angle compensation methodbased on coherency matrix have also confirmed the efficiency of the proposedcompensation methods.