基于形态学自适应分块的高分辨SAR多特征增强算法

Multi-feature enhancement algorithm for high resolution SAR based on morphological auto-blocking

传统基于■1范数正则化算子(least absolute shrinkage and selection operator,LASSO)模型的合成孔径雷达(synthetic aperture radar,SAR)压缩感知类稀疏成像算法易丢失弱散射点。基于扩展型组LASSO系列模型的算法虽可增强SAR结构特征以保留弱散射点,但由于其增强过程中本质上采用了欧式距离方法进行特征分块,致使分块较为“机械”,不能很好地提取目标结构,从而影响最终高分辨SAR成像质量。针对上述问题,提出一种基于形态学自适应分块的交替方向多乘子法(morphological auto-blocking alternating direction method of multipliers,MAB-ADMM)来实现高分辨SAR多特征表征。该算法通过建立基于形态学分块的■M/■F混合结构范数和■1稀疏范数来分别引入结构和稀疏先验,从而实现结构与稀疏多特征增强。由于采用了基于测地距离的形态学分块方式,MAB-ADMM算法能够更加有效地识别感兴趣的目标轮廓,从而提高结构增强的准确度和完整度。实验部分通过采用仿真复数据和实测SAR数据对所提算法和传统算法的成像结果进行定性对比,从而验证所提算法具有优越的多特征增强能力。此外,采用相变热力图对所提算法和传统算法的恢复能力进行定量对比,并利用MSTAR数据验证了所提算法可有效针对分类算法进行结果提升。

The conventional compressive sensing sparse imaging algorithm of synthetic aperture radar(SAR)based on the least absolute shrinkage and selection operator(LASSO)is easy to lose weak scatters.Although the imaging algorithm based on extended group LASSO is capable of strengthening weak scatters of structure of target of interests,it is inflexible to adaptively block the groups for enhancing the target feature,since only the Euclidean distance is adopted for the blocking operation,and the quality of the SAR imagery is poor.In this paper,a morphological auto-blocking alternating direction method of multipliers(MAB-ADMM)algorithm is proposed for the structural feature enhancement of high-resolution SAR imagery.A mixture of ■M/■F norm and an ■1 norm are introduced to represent the structural and sparse features,respectively,and a coordination process is carried out to enhance the features cooperatively.Because the proposed MAB-ADMM algorithm employs the morphological blocking scheme based on the geodesic distance,it is capable of capturing the contour of target of interests effectively,and the accuracy and integrity of the enhancement can be improved.In the experiment,both the simulated and raw SAR data are used for the verification of the proposed algorithm.Comparisons with the conventional algorithm are performed to show the superiority of the proposed algorithm.The phase transition analysis is carried out to evaluate the performance of the proposed algorithm quantitatively.Also,the MSTAR data set is employed to verify the performance of the proposed algorithm in preparing for the SAR target classification.