Research On Nonlocal Mean Filtering Algorithm For PolSAR Image

Polarimetric Synthetic Aperture Radar(PolSAR)uses different combinations of antenna polarization transceivers and receivers to measure ground objects.Compared with SAR,it can obtain more abundant information and has important research value and wide application prospects.However,due to the inherent defects of imaging mechanism,speckle noise will inevitably be introduced into PolSAR images,which not only leads to the decline of image visual quality,but also makes it difficult for subsequent classification,recognition and other process.It is always difficult to preserve the structure and polarization scattering characteristics of PolSAR image after speckle suppression.In order to solve this problem,based on the basic theory of PolSAR,this thesis summarizes several speckle suppression algorithms and their advantages and disadvantages.On this basis,two polarimetric SAR image denoising algorithms based on non-local mean denoising are proposed:(1)Dual-scale PolSAR image denoising algorithm based on non-local mean.In this method,strong point target detection algorithm is used to preserve the target with strong scattering characteristics in the image;pixels are classified by arithmetic-geometric mean ratio(AG),and a appropriate scale non-local mean filtering and smoothing coefficients are selected according to different filtering requirements,which can improve the edge preservation effect on the basis of guaranteeing the effect of speckle suppression.(2)In order to improve the adaptive degree of existing algorithms and the comprehensive denoising effect,an improved non-local mean denoising algorithm for PolSAR images is proposed based on the coefficient of variation.The algorithm combines the statistical characteristics of image sub-blocks and the polarization scattering characteristics of target points to filter the homogeneous pixels,which improves the accuracy of homogeneous pixel selection,and introduces the coefficient of variation to select the smoothing coefficient adaptively to achieve adaptive denoising in different regions.Comparative experiments with polarimetric SAR data collected by different systems,the results show that the proposed algorithms have better effect on the edge texture and polarization scattering characteristics of the image based on the effective suppression of speckle noise.