| The study of statistical characteristics of Synthetic Aperture Radar(SAR)images has important theoretical significance and application value for the development of SAR image interpretation algorithms.SAR image statistical modeling aims to explain the statistical characteristics of SAR image from the perspective of statistics,so as to reveal the scattering mechanism of scene targets in SAR image.In the field of SAR image interpretation algorithms,statistical model-based algorithms are most widely used.With the continuous development of SAR technology,the resolution of SAR image is gradually improved.On the one hand,high-resolution SAR image contains more details and texture information of ground objects,on the other hand,it also makes the statistical characteristics of its image more complex,making it difficult for the classical SAR statistical model to accurately describe the statistical characteristics of high-resolution SAR image.In order to further improve the level of SAR image interpretation,the research on SAR image statistical modeling and its segmentation algorithm is carried out in this paper.The main research contents are as follows:(1)An image statistical model based on the generalized Gamma distribution is proposed to solve the problems of complex texture structure and statistical characteristics of high-resolution SAR images.Under the framework of the product model of SAR image,the speckles are assumed that obey the unit mean Gamma distribution,and the Radar Cross Section(RCS)is accurately described by the generalized Gamma distribution.Then,the generalized compound Gamma distribution is constructed,and the parameter estimation formula based on the Method of Log-Cumulants(Mo LC)is deduced.(2)A regionalized SAR image segmentation algorithm combining generalized Gamma distribution and Markov Random Field(MRF)is proposed to improve the segmentation accuracy of high-resolution SAR images.Firstly,the SAR image is divided into several subregions by Voronoi partitioning,and each subregion is given the same label.Secondly,the statistical distribution model among pixels in the sub-region is established by using the statistical model based on the generalized Gamma distribution,and the interaction relationship between label fields is described by using the MRF model.Finally,the image segmentation model was established according to Bayes theorem,and the Markov Chain Monte Carlo(MCMC)algorithm was designed to solve the segmentation model and the optimal model parameters,and the optimal image segmentation was obtained by maximizing the posterior probability.(3)Verify the effectiveness of the proposed model and segmentation algorithm.High-resolution SAR images of different polarization modes and different scenes were used to conduct statistical modeling and image segmentation experiments,and the experimental results were analyzed qualitatively and quantitatively.Experimental results show that the proposed model has higher modeling accuracy compared with the classical distribution model,and the proposed segmentation algorithm effectively improves the segmentation quality compared with the conventional SAR image segmentation algorithm. |
