Study Of Information Preprocessing And Target Decomposition Methods For Polarimetric SAR

Polarimetric SAR(PolSAR)collects full polarimetric scattering information from the target at the mean time of imaging,which is one of the important research topics in microwave remote sensing technology.This dissertation works on the preprocessing of polarimetric information(PolSAR image speckle filtering and target deorientation)and target decomposition(model-based decomposition and eigenvalue/eiegnvector based decomposition).The major contributions are summarized as follows:1.Study on speckle filtering for PolSAR image.We proposed an PolSAR speckle filter which can simutaneously preserve the polarimetric scattering mechanism and target edge information of the target.The polarimetric similarity parameter and power similarity parameter between the center pixel and the surrounding pixels in a sliding window are tested to measure their similarity degree of polarimetric scattering mechanism and power information,respectively.Then the thresholding methods are applied,and the surrounding pixels whose polarimetric scattering mechanism and power information consistent with those of the center pixel are selected to join the filtering.The power information directly reflects target edge information.2.Study on the target deorientation method.The existing deorientation method does not work well for targets with high entropy,such as for the largely oriented buildings.Within the resolution cell of these targets,there are a number of scatterers with different orientation angles(OA),and one OA cannot effectively describes these scatterers for the existing method.Based on the decomposition method,we proposed to decompose the distributed target into three single targets which have different OAs,and the three single targets represent all the scatterers.The three single targets are differently deoriented,and thus an improved deorientation result is achieved.3.Study on the model-based decomposition methods for PolSAR.For the nonnegative eigenvalue decomposition,the form of the remainder matrix is not as same as the form of models of surface scattering component and double-bounce scattering component,and the surface scattering component and double-bounce scattering component are not derived by model-based methods.We propose to implement the deorientation and further helicity angle compensations for the remainder matrix,and then the form of the compensated remainder matrix is just as same as the model forms of surface scattering component and double-bounce scattering component.Lastly the surface scattering component and double-bounce scattering component are derived by modelbased methods.4.Study on the eigenvalue/eigenvector-based decomposition methods for PolSAR.A sufficient and lossless eigenvalue/eigenvector based target decomposition method for coherency matrix T is proposed.Based on the multiplicative decompsotion framework,nine physically explicit parameters are decomposed from the coherency matrix T which has nine degrees of freedom,and the dominant eigenvector is modelled by target scattering vector model(TSVM).An improved target fine classification scheme is also proposed,and real PolSAR data are processed to validate the proposed target decomposition method and classification method.