系统工程与电子技术 ›› 2022, Vol. 44 ›› Issue (3): 808-818.doi: 10.12305/j.issn.1001-506X.2022.03.13

• 传感器与信号处理 • 上一篇    下一篇

高分辨SAR目标成像方向性结构特征增强

杨磊*, 张苏, 盖明慧, 方澄   

  1. 中国民航大学天津市智能信号与图像处理重点实验室, 天津 300300
  • 收稿日期:2020-10-14 出版日期:2022-03-01 发布日期:2022-03-10
  • 通讯作者: 杨磊
  • 作者简介:杨磊(1984—), 男, 副教授, 博士, 主要研究方向为高分辨SAR成像及机器学习理论应用|张苏(1996—), 女, 硕士研究生, 主要研究方向为高分辨SAR成像及优化学习理论|盖明慧(1997—), 女, 硕士研究生, 主要研究方向为高分辨SAR成像及优化学习理论|方澄(1980—), 男, 讲师, 博士, 主要研究方向为深度学习及目标异常检测
  • 基金资助:
    国家自然科学基金(61601470);天津市自然科学基金(16JCYBJC41200)

High-resolution SAR imagery with enhancement of directional structure feature

Lei YANG*, Su ZHANG, Minghui GAI, Cheng FANG   

  1. Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China
  • Received:2020-10-14 Online:2022-03-01 Published:2022-03-10
  • Contact: Lei YANG

摘要:

针对传统稀疏特征增强的方式仅能完成对目标场景中特显点的增强, 对复杂的目标结构特征无能为力的问题, 考虑目标细节特征的复杂性, 提出方向性结构全变分(directional total structure variation, DTSV)正则子进行结构先验表征, 实现对成像目标复杂结构特征任意梯度变化的拟合, 进而实现对结构特征的高精度正则优化处理。首先, 在交替方向多乘子方法(alternating direction method of multipliers, ADMM)的协同优化框架下实现DTSV正则优化求解(DTSV-ADMM), 利用该框架提供的对偶上升思想可有效提升迭代优化算法的收敛性能。其次, 基于ADMM框架提供的多变量“分解-调和”机理, 通过建立分裂变量组可以实现多个正则项的协同优化增强。然后, 进一步引入${\ell _1}$范数对成像目标稀疏特征进行表征, 并在协同优化框架下实现对方向性结构特征和稀疏特征的稳健计算, 有效减小多特征优化存在的“误差传播”问题。最后, 通过近端算子对特征进行解析计算, 获得对应特征的闭合解析解, 进一步提升算法运算稳健性和计算效率。实验证明了所提算法相比传统方法的优越性。

关键词: 合成孔径雷达, 方向性结构全变分, 交替方向乘子法, 多特征增强, 近端算子

Abstract:

Aiming at the problem that the traditional sparse feature enhancement method can only enhance the salient points in the target scene and cannot do anything about the complex target structure features, considering the complexity of the target detail features, a directional total structure variation (DTSV) regularizer is proposed for a priori structural feature. The fitting of arbitrary gradient change of complex structural features of imaging targets is realized, and then the high-precision regularization optimization processing of structural features is realized. Firstly, DTSV-alternating direction method of multipliers (DTSV-ADMM) is implemented under the collaborative optimization framework of ADMM. The dual ascent idea provided by this framework can effectively improve the convergence performance of iterative optimization algorithm. Secondly, based on the multivariable "decomposition-coordination" mechanism provided by ADMM framework, the cooperative optimization and enhancement of multiple regular terms can be realized by establishing split variable groups. Thirdly, the ${\ell _1}$ norm is further introduced to characterize the sparse features of imaging targets, and the robust calculation of directional structural features and sparse features is realized under the framework of collaborative optimization, so as to effectively reduce the problem of "error propagation" in multi feature optimization. Finally, the features are analytically calculated by the proximity operator to obtain the closed analytical solution of the corresponding features, which further improves the computational robustness and computational efficiency of the algorithm. Experiments show that the proposed algorithm is superior to the conventional method.

Key words: synthetic aperture radar (SAR), directional total structure variation (DTSV), alternating direction method of multipliers (ADMM), multi-feature enhancement, proximity operator

中图分类号: