Polarimetric And Spatial Feature Extraction And Application Of PolSAR Image Based On Convolution Sparse Representation

Polarimetric SAR(PolSAR)not only has the advantages of SAR system,but also can transmit and receive echo signals of different polarization modes to obtain more abundant ground object scattering information.Therefore,it has important application value in many fields such as agriculture,forestry,ocean,urban construction and military reconnaissance.However,with the continuous update of PolSAR system,the resolution of obtained image is improved constantly,the target is no longer represented as an isolated point or composed of a few pixels in the image,but as a region with complete geometric structure,minute details and texture information.So in PolSAR image target detection and classification,it is unreasonable to consider only the polarimetric information of a single pixel,the spatial environment information around the target pixel also needs to be fully considered.Therefore,it is of great significance to study the methods that can effectively combine polarimetric information and spatial information of PolSAR images.The convolution sparse representation theory can realize the connection between adjacent information in the image space through convolution,retain the structural and texture information of the image as possible,avoid data redundancy and loss of image structure information in the sparse representation process.So based on PolSAR image polarimetric feature extraction,this paper uses the convolution sparse representation theory to realize the combination of polarimetric and spatial information,and research is carried out on the application of PolSAR image target detection and ground classification.The specific research contents are as follows.Firstly,a two-stage target decomposition model based on scattering symmetry is proposed to solve the problem that the scattering symmetry hypothesis is not fully considered in the traditional incoherent target decomposition method.The coherent matrix is decomposed into scattering symmetric component and scattering asymmetric component,and further the scattering symmetric component is decomposed into surface,double and volume scattering,the asymmetric component is decomposed into helix,wire,oriented dipole and compound dipole scattering as well as compound asymmetric scattering,and the volume scattering model is improved,suppressing the overestimation of volume scattering.The method effectively utilizes the scattering symmetry of PolSAR data,and can describe the scattering characteristics of the target in more detail.The polarimetric characteristics obtained by decomposition can be used for target detection and image classification applications in subsequent PolSAR images.Secondly,in order to solve the problem that the spatial information is difficult to effectively use for man-made target detection in high-resolution PolSAR image,a joint polarimetric and spatial target detection method based on convolution sparse representation and background constraint is proposed.The polarimetric feature of PolSAR image is modeled by convolution sparse representation,and the background feature is modeled and combined with regularization constraint,which can realize the combination of polarimetric features and spatial information.Furthermore,for the difficulty of background estimation,a joint polarimetric and spatial target detection model based on local convolution sparse representation is proposed.This method models the polarimetric features of the target based on the convolution sparse representation,uses the local convolution strategy to construct the target dictionaries and feature responses,and separates the target from the PolSAR image to achieve target detection by iterative,avoiding the interference from other objects when using global convolution sparse representation.Finally,for the problem that the polarimetric and spatial feature extraction of objects in high-resolution PolSAR images are not closely combined,a joint polarimetric and spatial feature extraction method based on local convolution sparse representation is proposed.In this method,the traditional incoherent target decomposition model is represented by convolution sparse representation at the spatial level of the image,uses local strategy to construct feature responses and dictionary filters,uses the feature responses to represent the joint polarimetric and spatial features of PolSAR image,and realizes the integration of polarimetric and spatial information.Also,it can avoid the confusion between different objects in the global convolution process.Finally,the PolSAR image classification is used to verify the validity of the joint polarimetric and spatial features.