Research On The Methods Of Despeckling And Classification Based On Non-local High Resolution SAR Image

High-resolution SAR image could show the features more clearly than the lowresolution SAR image.However,the subtle differences between the features of the same kind influence the imaging with diverse reflections.Thus,the traditional image processing methods of moderate or low resolution SAR image are no longer suited to high resolution SAR image.In the high resolution SAR image,a lot of information details make the image with strong redundancy,that is,the image has more similar information but there is no adjacent relationship in the spatial distance with non-local characteristics.In this paper,based on the self-similarity in high resolution SAR image,study on the high resolution SAR image processing,focused on the processing of despeckling and classification.The main research work and innovation of this paper are as follows:(1)High resolution SAR image speckle reduction method based on non-local weighted nuclear norm minimization is improved.Based on study of the non-local speckle reduction method and the nuclear norm minimization,improve the despeckling method that a simple non-local kernel-norm minimization method for speckle reduction has problems in strong speckle noise.Thus,after the pre-despeckling by PCA,the method of minimizing the nuclear norm minimization is adopted for better structure preserving.The experimental results show that the proposed despeckling method can remove the speckle noise while preserving the structure information of the original image,and has the ideal despeckling result in simulation image and real SAR image.(2)High resolution SAR image speckle reduction method based on non-local clustering nuclear norm minimization is proposed.Aiming at the problem that the nonlocal information is not closely related with the filtering method,the low-rank matrix is constrained by the non-local information obtained by clustering.The clustering-based nuclear norm minimization model is proposed and solved by the iterative reweighted and regularized nuclear norm minimization,then used in SAR image despeckling.The experimental results show that the SAR image despeckling method with non-local clustering-based nuclear norm minimization has better structure preserving ability,and clearer edge profile on the edge processing,which is proved by the simulation image and the real SAR image.(3)A high-resolution SAR image unsupervised classification method based on non-local super-pixels is proposed.The method of fuzzy C-means clustering is improved,which is easy to be disturbed by the noise points in high resolution SAR image.Combine the non-local information obtained by non-local based nuclear norm minimization despeckling method with local information obtained by super-pixel segmentation to improve the classification accuracy.Finally,a method of unsupervised high resolution SAR image classification based on non-local super pixels is presented.The experimental results show that the classification method based on non-local super-pixel has high classification precision without the influence of ground object distribution on both simulation images and real SAR images.