Research On Despecking And Segmentation Algorithms Of SAR Image

Synthetic Aperture Radar(SAR)is an active microwave imaging system with allday and all-weather characteristics.At present,as a significant means of acquiring geospatial information,SAR imaging technology has been widely applied in military investigation,scientific research and resource exploration fields.However,the interpretation of SAR images is difficult due to a large amount of speckle in the image.Therefore,it is of great significance to take effective measures to suppress speckle for the application of SAR images.Meanwhile,SAR image segmentation is a critical part in SAR image preprocessing,which directly affects the accuracy of subsequent image interpretation.In this dissertation,the SAR speckle filtering technology and image segmentation algorithm are studied.The main contents are summarized as follows:(1)The characterization of polarimetric SAR image,the common target decomposition algorithm and the statistical characteristics of speckle are summarized,which lays a foundation for the research of PolSAR despecking.(2)A polarimetric SAR despeckling algorithm based on polarimetric target decomposition and NL-Lee filter is proposed.Firstly,this paper introduces a polarization similarity parameter to improve the accuracy of pixel similar set selection.Moreover,NLLee filter is combined with the scattering model of PolSAR to construct the framework.And a weighted acceleration algorithm is proposed to eliminate redundant operations.Finally,the experiment results of real PolSAR images demonstrate that its rapidity and competitive general performance of speckle reduction and edge preservation.(3)A fuzzy C-means clustering algorithm based on superpixel and sparse representation(SSR_FCM)is proposed,which introduces sparse representation theory and processes the image at superpixel level to improve the segmentation performance as well as efficiency.At first,the PILS algorithm is utilized to generate superpixels as the basic unit of subsequent steps.Futhermore,this paper applies the sparse representation theory with correction preprocessing to extract distinguishing features of superpixels.Then the two types of features are clustered independently and simultaneously.The experiments on both simulated and real SAR images indicate that the algorithm proposed by this dissertation has competitive segmentation performance.(4)A robust non-local FCM clustering algorithm with comentropy and between-cluster scatter matrix(NCBS_FCM)is presented.This paper introduces an adaptive similarity measure on the basis of the similarity calculation method in the non-local mean algorithm in order to enhance the robustness against speckle noise.Moreover,the comentropy of the gray-level histogram is utilized to calculate the value of the adaptive weighting parameter,which represents the weight of the nonlocal spatial information term.And the between-cluster scatter term is introduced into the objective function simultaneously.Experiment results on both simulated and real SAR images show that the proposed algorithm has great performance in terms of speckle suppression and detail preservation.Compared with other variations of FCM,the segmentation accuracy and adjusted rand index of NCBS_FCM increase 17.1% and 30.6% respectively,which indicate that NCBS_FCM outperforms other variations of FCM by a significant margin.