Regularization Methods For Feature Enhancement Of SAR Images

As a high-resolution microwave imaging radar,synthetic aperture radar(SAR)allows the observation of broad expanses day and night as well as without the effects of weathers.These characteristics make it wildly used in numerous fields such as resource exploration and military reconnaissance.Feature of SAR images is one of the important factors to measure the quality which largely determines the distinctiveness of the SAR images.The strength of regularization approach is that the prior information can be directly integrated into an objective function for feature enhancement of SAR images.Therefore,the prior information of features can be modeled to enable the accurate numerical solution.It's our goal in this paper to develop regularization models for feature enhancement of SAR images based on its prior information and design fast optimization algorithms.The main contributions are as follows:Firstly,the feature degradation model of SAR images is proposed.Starting with the transfer function of SAR imaging system,we analyze the phase errors resulted from non-ideal imaging conditions and the relations of error propagation and study the mechanism of feature degradation and propose the prior model of classic feature degradations.We extract point features and area feature such as dihedral angle for model identification and parameter estimation,and then determine the feature degradation model as prior information to enhance the features of SAR images.Secondly,regularization methods based on sparsity and partial differential equation(PDE)for feature enhancement of SAR images is proposed.Based on the sparse prior of SAR images,we introduce PDE penalty with ROA edge detector to sparse regularization model and adopt alternating optimization algorithm for the optimal subproblems.The results of numerical experiments demonstrate that the new method can overcome the deficiency of not protecting the structure of area targets and enhance the feature of strong scattering point and line target,which expand the application range of feature enhancement methods.Thirdly,regularization methods based on nonlocal low rank for feature enhancement of SAR images is proposed.For SAR images with periodic and similar structures,we combine nonlocal method with minimization theory of matrix rank and develop a new regularization method for feature enhancement of SAR images with nonlocal low rank constraint.The alternating optimization minimization algorithm is adopted for the optimization problem.The results of numerical experiments show the new method improves the performance of feature enhancement for these nonlocal and low-rank SAR images.