基于稀疏贝叶斯方法的THz-SAR振动目标成像研究

Target vibration conveys favorable information for recognition, and imgaing vibrating targets with synthetic aperture radar (SAR) bears considerable significance to battlefield reconnaissance, precision guidance, traffic monitoring and antiterrorism. However, vibration has minor amplitude and Doppler, and it will induce complex modulation as well as image defocusing, hence difficult for detection. Therefore, we at first propose to enhance micro-Doppler by incorporating THz techniques, and then build in the return domain a measurement model based on the hybrid vibration-scattering center. Thus image reconstruction is reformulated as a Fredholm equation inverse problem for finding vibration parameters and scattering coefficients. In the end, target sparse priors are exploited and used for fast and global optimization in the Bayesian theory framework. The idea avoids conventional difficult procedures, e.g. clutter suppression and range cell migration correction, directly realizes optimal estimation of vibration and scattering parameters, and can obtain the distribution of scattering coefficients in both space and vibration-parameter dimensions, hence mitigating blurs in original SAR images and enhancing target features. This project will not only provide supports for SAR sensing of special targets, but also can promote the development of sparse Bayesian application studies.

目标振动蕴含着对识别极为有利的信息，SAR地面振动目标成像对战场侦察、精确制导、交通监视、维稳反恐等具有重要意义，但存在两个难题：一是振动幅度小，多普勒效应弱；二是振动对多普勒调制机理复杂，成像难度大。对此，本项目首先引入THz技术以增强微多普勒效应，然后基于振动-散射中心混合模型从回波域构建THz-SAR观测模型，将成像转化关于振动参数和散射系数的Fredholm方程求解逆问题。最后，在贝叶斯框架下开采目标稀疏先验，研究先验支持下的快速全局优化算法。这一思路避免了传统杂波抑制、距离徙动校正等复杂处理环节，能够实现目标振动和散射参数的最优估计，获取目标散射系数在空间位置和振动参数维度上的分布即超立方体图像，达到SAR图像中模糊消除和特征凸显的效果。项目将为SAR特种目标探测提供理论支撑，同时推动稀疏贝叶斯应用研究的发展。

本项目以战场侦察为研究背景，围绕THz-SAR振动目标成像问题，重点突破了目标振动特性测量与建模、太赫兹频段车辆目标RCS计算测量以及基于稀疏贝叶斯方法的振动目标成像等技术，取得的成果主要包括：对振动目标进行特性测量和分析，得到车辆振动的基本特征；计算并测量了太赫兹频段典型标准体和车辆模型的散射特性，为回波建模提供数据基础；建立了基于第一类Fredholm方程的振动目标回波模型，并分析得到太赫兹频段目标振动对SAR图像的影响规律；研究了稀疏贝叶斯模型及其求解算法，提出了基于方差成分扩张压缩方法的振动目标成像和参数估计方法。以项目研究成果为基础，出版专著1部，获2项国防专利授权，发表学术论文24篇，SCI检索13篇，EI检索23篇。