Research On Target Detection Methods Of Polarized Hyperspectral Images Based On Tensor And Sparse Representation

The polarized hyperspectral images(PHSI)can be seen as the combination of hyperspectral images(HSI)and polarized images,and they don't only posses all the information of HSI,but also has the polarization information.Polarization can enhance the contrast of images,eliminate the influence of interfering signal,which improves the target detection rate.Therefore,the research on target detection methods of PHSI conducted in this thesis is as follows.Firstly,the basic concepts of Stokes vector and its application in polarized hyperspectral image are introduced.For the intensity,polarized degree and polarized angle extracted from Stokes vector,the RX detection algorithm of unknown background and unknown target,CEM detecton algorithm of unknown background and known target and SSP detection algorithm of known background and known target are applied to these components of Stokes.Finally,based on the three models,these detection results are fused by fuzzy integral,and the improvements of the detection rate are achieved.thinking about the multi-dimensional information of PHSI and the tensor can ensure the integrity of the data information,the concept of tensor representation is introduced in this thesis,then the tensor decomposition is used to extract spectrum and polarization features of PHSI,and on this basis,a fourth-order tensor matching filter and a target detection algorithm based on tensor CP decomposition reconstruction,which not only realize the combined ultilization of multi-dimensional information,but also overcome the sensitivity of the detection algorithm to noise to a certain extent and get the better detection results than the classical methods.The polarization hyperspectral images contain rich multi-dimensional information,and in view of the fact that sparse representation not only doesn't need to consider the distribution of data,but also can realize the effective utilization of spatial information,hence,in order to take advantage of space information,we have combined tensor and sparse and proposed a sparse representation target detection algorithm based on tensor decomposition reconstruction,which realizes the combined utilization of polarized,spectral and spatial information,and reduces the false alarm,improves the target detection rate of PHSI further.