Research On Variational Segmentation Of Remote Sensing Data For Fusion Image Restoration

Synthetic aperture radar(SAR)is an important method for acquiring remote sensing images.The images produced by SAR contain rich information,but there is also a large amount of multiplicative speckle noise,which increases the degree of image blur.Resulting in the failure to obtain ideal results in the subsequent SAR image segmentation.Therefore,SAR image segmentation has always been a hot issue for researchers.Among the various image segmentation methods that have been proposed,the variational segmentation method based on variational and partial differential equations,with its unique advantages,has become one of the most effective methods for SAR image segmentation.However,the existing variational segmentation model is suitable for smooth or additive noise images.For SAR images with multiplicative speckle noise,it is necessary to convert the multiplicative noise into additive noise first,which has low time efficiency and poor model convergence,resulting in segmentation results It is difficult to meet the demand.This paper combines the multiplicative noise recovery model and the image segmentation model to construct a new SAR image segmentation method,and converts a small number of calibrated Landmarks into constraint points to ensure that the curve evolution can pass through the Landmark points,so as to enhance the method's segmentation ability.The main content and innovation are as follows:(1)A SAR image variational segmentation model incorporating multiplicative noise removal is proposed.After fully considering the statistical characteristics of speckle noise,based on the gamma distribution,the data items for SAR image speckle noise are designed and introduced into SAR image segmentation.Through the improvement of the multiphase variational level set Chan-Vese model,so that the proposed model can combine denoising and segmentation,not only has better denoising effect,but also can obtain more accurate segmentation results.(2)A multiphase segmentation model of remote sensing images based on Landmark points is proposed.Landmark points are a priori image features for various computer vision tasks.This paper uses the level set framework to convert the landmark points of the image into simplex constraints and incorporate them into the variational segmentation model,so that the level set curve evolution process can pass all priors Landmark points to improve segmentation performance.(3)Design the fast Alternating Direction Method of Multipliers(ADMM)algorithm for the variational segmentation model.By introducing a series of auxiliary variables,combined with the augmented Lagrangian method,the complex energy functional is transformed into a number of simple sub-optimization problems,the generalized soft threshold formula is used to solve the algebraic equation,and the projection method is used to satisfy the relevant inequality constraints,so as to realize the fast solution of the proposed segmentation method.Finally,the SAR image segmentation experiment is carried out.The experimental results show that the image segmentation method proposed in this paper can obtain more accurate segmentation results.