雷达目标检测的信息几何方法

Target detection under the condition of complex clutter background and low signal-to-clutter ratio (SCR) is a difficult problem in radar signal processing. The difficulties lie in: 1) The recognition of complex clutter environment is very difficult. Existing clutter models are difficult to characterize the intrinsic information of clutter data, and the detection performance suffers from a sharp decline. 2) The information difference between statistical models cannot be fully explored by the existing detection methods, which brings the loss of detection performance. 3) It is difficult to improve the detection SCR when the number of data samples available are not enough. Consequently, the improvement of detection performance is greatly restrained. In order to break through above difficulties, it is urgent to explore new theory and methods for radar target detection.. With respect to the above problems, this project exploits the modern differential geometry method on statistical manifold of probability distribution families in the framework of information geometry, to establish the relationship between the geometric structures of statistical manifold and target detection problem. This methodology can provide a new viewpoint to study detection problems and algorithms. According to the three fundamental and scientific problems mentioned above, such as recognition of clutter characteristics and its intrinsic information representation, the measurement of information difference between statistical models, and the information-geometric method of target detection on statistical manifold, and so on, the underlying project will carry out several research achievements. These achievements will put forward a number of new ideas and methods for radar target detection, and promote the technology development of radar target detection.

复杂背景及低信杂比条件下的目标检测是雷达信号处理面临的一个难题，受到学术界的广泛关注。其难点在于：探测环境复杂，背景杂波认知难度大，现有杂波建模方法难以表征杂波数据的内蕴信息，导致检测性能的下降；常规检测方法难以充分挖掘统计模型间的信息差异，带来了检测性能的损失；在有效检测样本数较少的情况下，检测信杂比难以提高，检测性能受到较大制约。上述难点问题的突破，迫切需要探索研究雷达目标检测的新理论与新方法。. 本项目针对上述难点问题，以信息几何理论为基础，在概率分布流形上采用现代微分几何方法，建立流形的几何结构与目标检测问题之间的联系，为检测问题与检测方法的研究提供一个全新的视角。针对杂波特性认知与内蕴信息表达、统计模型信息差异高效度量、目标检测的信息几何方法等基础性、科学性问题展开深入研究，提出雷达目标检测的新思路与新方法，促进雷达目标检测技术的发展。

项目针对复杂环境下雷达弱小目标检测难题，开展基于信息几何的雷达目标检测方法研究。围绕杂波特性认知与内蕴信息表达、统计模型信息差异高效度量、目标检测的信息几何方法与实验验证等三项研究工作，提出了基于信息几何的雷达目标检测新方法，为低信杂比目标检测提供了技术途径。.在杂波特性认知与内蕴信息表达方面，基于雷达实测数据集，分析了杂波与目标在时间和空间维度上的相关性差异，提出了基于统计流形和正定矩阵流形的杂波内蕴信息建模方法，为实现信号流形上目标检测奠定基础。.在统计模型信息差异高效度量方面，研究了流形测地线距离和信息散度两类差异性度量方法，比较了7种差异性度量的几何性质，从稳健性和仿射不变性两个方面准确度量目标与杂波统计特性差异性。.在目标检测的信息几何方法与实验验证方面，提出了基于度量学习的低信杂比目标检测方法和基于仿射变换的非均匀杂波自适应目标检测方法，采用加拿大IPIX雷达海杂波测量数据、雷达学报“X波段雷达对海探测试验数据”对所提方法进行了性能验证。相比传统检测方法，信息几何检测方法有效提升了复杂环境下雷达弱小目标检测性能。研究成果预期可应用于复杂环境下无人机、车辆等目标的有效探测，为雷达探测性能提升提供技术支撑。.在该项目支持下，项目组在国内外高水平期刊和会议发表论文23篇，SCI检索14篇，EI检索8篇，中文核心期刊论文1篇。其中，在IEEE TSP、IEEE TGRS上发表长文3篇（中科院1区3篇），在IEEE GRSL、IEEE SPL、IEEE CL上发表论文6篇（中科院2区5篇）。累计18人次参加国内外学术交流，项目负责人在国际期刊《Sensors》及国际会议IEEE ICCT 2022组织2期信息几何研究专刊，并担任专刊主编和分会主席。授权国家发明专利5项，申请4项。项目共培养博士/硕士研究生6名。项目负责人分别于2021年和2022年获湖南省自然科学一等奖1项（排名3）和军队科技进步一等奖1项（排名5）。项目相关成果“信息几何雷达目标探测技术”入选国防科技大学2022年度十大科技进展。