杨磊^,,
夏亚波,
毛欣瑶,
廖仙华,
方澄,
高洁
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
基金项目:中央高校基本科研业务费专项资金(3122018C005, 3122014C009)，国家自然科学基金(61601470)，天津市自然科学基金(16JCYBJC41200)

详细信息

作者简介:杨磊：男，1984年生，副教授，研究方向为高分辨SAR成像及机器学习理论应用
夏亚波：男，1991年生，硕士生，研究方向为高分辨SAR成像及统计采样技术应用
毛欣瑶：女，1993年生，硕士生，研究方向为高分辨SAR成像及微多普勒特征增强
廖仙华：男，1993年生，硕士生，研究方向为高分辨SAR成像及统计采样技术应用
方澄：男，1980年生，讲师，研究方向为深度学习及目标异常检测
高洁：女，1984年生，讲师，研究方向为多源数据融合

通讯作者:杨磊　yanglei840626@163.com

中图分类号:TN957.52

计量

文章访问数:1002
HTML全文浏览量:258
PDF下载量:67
被引次数:0

出版历程

收稿日期:2020-04-21
修回日期:2020-08-08
网络出版日期:2020-08-13
刊出日期:2021-03-22

Sparse ISAR Imaging Algorithm Based on Bayesian-Lasso

Lei YANG^,,
Yabo XIA,
Xinyao MAO,
Xianhua LIAO,
Cheng FANG,
Jie GAO
Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China
Funds:The Fundamental Research Funds for Central Universities of Ministry of Education of China (3122018C005, 3122014C009), The National Natural Science Foundation of China (61601470), The Natural Science Foundation of Tianjin (16JCYBJC41200)


摘要
摘要:逆合成孔径雷达(ISAR)目标回波具有明显的稀疏特征，传统的凸优化稀疏ISAR成像算法涉及繁琐的正则项系数调整，严重限制了超分辨成像的精度及便捷程度。针对此问题，该文面向非约束Lasso正则化模型，建立分层贝叶斯概率模型，将非约束的$ {\ell _1}$范数正则化问题等效转化成稀疏拉普拉斯先验建模问题，并在分层贝叶斯Lasso模型中建立正则项系数依赖的概率分布，从而为实现完全自动化参数调整提供便利条件。考虑到目标稀疏散射特征和多超参数的高维统计特性，该文应用吉布斯(Gibbs)随机采样方法，实现对ISAR目标稀疏特征的求解，并同步获取包括正则项系数在内的多参数估计。基于该文研究方法可实现全部参数均通过数据学习获得，从而有效避免繁琐的参数调整过程，提升算法的自动化程度。仿真及实测数据均可证明该方法的有效性和优越性。
关键词:逆合成孔径雷达/
贝叶斯Lasso/
分层贝叶斯/
吉布斯采样
Abstract:Due to the echoes of the Inverse Synthetic Aperture Radar (ISAR) imagery are spatially sparse, the conventional convex optimization for the sparse image recovery involves tedious adjustment for the regularization parameter, which seriously limits the accuracy and the convenience of the image formation. In this paper, the unconstrained least absolute shrinkage and selection operator (Lasso) model is introduced for the $ {\ell _1}$ regularization problem, and it is equivalently transformed into sparse Bayesian inference under the Laplacian prior. More specifically, a hierarchical Bayesian model is established. In such cases, multiple hyper-parameters with multi-level conditional probability distribution are introduced. Due to the equivalent transformation, the manual choice of the regularization parameter can be replaced by automatic determination under the hierarchical Bayesian model, which provides convenience of fully conditional probability adjustment. Considering the high dimensions of sparse image responses and multiple hyper-parameters, the Gibbs sampler is adopted, where the Bayesian posterior of the ISAR image and high-dimensional hyper-parameters can be solved with fully confidence. Based on the research in this paper, all parameters can be attained by data, therefore tedious parameter adjustment can be avoided, and the automation level of the algorithm can be improved. The effectiveness and superiority of this method are proved by both simulated and measured data experiments.
Key words:Inverse Synthetic Aperture Radar(ISAR)/
Bayesian Lasso/
Hierarchical Bayesian/
Gibbs sample



PDF全文下载地址:

https://jeit.ac.cn/article/exportPdf?id=06fcf763-c50c-442b-920d-ba0225f11d86