可变先验贝叶斯学习稀疏SAR成像  

Sparse SAR imaging based on varying prior Bayes learning

作　　者：沈笑云 廖仙华[1,2] 孙卫天 夏亚波 杨磊 SHEN Xiaoyun;LIAO Xianhua;SUN Weitian;XIA Yabo;YANG Lei(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;Tianjin Key Laboratory for Advanced Signal Processing,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300 [2]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

出　　处：《系统工程与电子技术》2021年第7期1781-1790,共10页Systems Engineering and Electronics

基　　金：国家自然科学基金(61601470);天津市自然科学基金(16JCYBJC41200)资助课题。

摘　　要：针对合成孔径雷达(synthetic aperture radar,SAR)在稀疏成像中,传统贝叶斯机器学习算法存在先验固化、成像结果容易过拟合等问题。提出一种可变成像先验贝叶斯(varying imaging prior Bayes,VIP-Bayes)学习稀疏SAR成像算法。首先,引入可动态灵活表征目标散射特征的广义高斯分布先验。然后,在贝叶斯推理框架下进行分层建模,后验分布推导。最后,针对常规吉布斯采样算法无法采样复杂后验分布的问题,引入哈密顿蒙特卡罗(Hamiltonian Monte Carlo,HMC)采样算法进行求解。另外,考虑到HMC算法对非平滑后验分布无法采样,因此引入近端算子,进行近端梯度近似,提出近端-HMC(proximal-HMC,P-HMC)算法。P-HMC算法可有效解决非平滑后验采样问题。因而可实现VIP-Bayes稀疏成像。通过仿真数据进行算法有效性验证,选取SAR实测数据与多种算法进行成像对比实验,利用相变热力图对算法成像性能进行定量分析,验证了所提算法的实用性和优越性。Aiming at the problem of prior solidification and over-fitting imaging results easily of traditional Bayesian learning imaging algorithms in sparse synthetic aperture radar(SAR)imagery,a varying imaging prior Bayes(VIP-Bayes)learning algorithm is proposed.Firstly,the generalized Gaussian distribution variable dynamic prior is introduced,which can realize the dynamic and flexible representation of the target scattering prior knowledge.Then,in the framework of Bayesian inference,a hierarchical Bayesian model is introduced to derive the posterior distribution.Finally,aiming at the problerm,which the conventional Gibbs algorithm cannot sample the obtained complicated posterior distribution,the Hamiltonian Monte Carlo(HMC)sampling algorithm is introduced to solve the problem.In addition,considering the HMC algorithm can’t sample for non-smooth posterior distributions.This paper introduces a proximal operator to approximate the gradient,and obtains the proximal-HMC(P-HMC)sampling algorithm.P-HMC algorithm can effectively solve the problem of non-smooth posterior sampling,for this reason,it can realize sparse imaging based on VIP-Bayes learning.The effectiveness of the algorithm is verified by using the simulation data,then select SAR raw data to conduct a variety of algorithm imaging comparison experiments.Finally,phase transition diagram is applied to quantitatively analyze the algorithm imaging performance.In a word,all above experiments verifies the practicability and superiority of the proposed algorithm.

关 键 词：合成孔径雷达 贝叶斯学习 哈密顿蒙特卡罗算法 广义高斯分布 

分 类 号：TN957[电子电信—信号与信息处理]