微波凝视关联成像目标特性建模与检测识别新方法研究

Microwave staring correlated imaging (MSCI) technology was proposed based on the concept of optical correlated imaging, and attracts more and more attention in recent years. It has realized high-spatial and high-temporal resolution microwave imaging onboard stationary platforms through the spatial and temporal two dimensional stochastic radiation field. However, most studies are only focused on the system development and image correlated reconstruction, while there is little research on its basic application theory under the specific imaging mechanization. This proposal firstly studies the scatterering characteristics of distributed targets under the microwave staring correlated imaging framework. In order to solve the signal coupling among the radiation antennas and the dispersion effects induced by the distributed targets, scattering characteristic model suitable for the distributed targets is developed through the antenna coupling analysis and finite difference time domain method of dispersion medium. Imaging simulation method is also developed. Secondly, in order to solve the problem that the traditional image feature description model, which originally constructed based on SAR image, is not applicable to describe the staring correlated images, sparse representation method for typical linear targets is developed based on the redundant dictionary, and related target detection and recognition models are developed. Thirdly, taking the advantage that the radiation field could be flexibly controlled and multi resolution image can be reconstructed, new application methods which integrate both the image reconstruction process and target recognition process are proposed. These studies will lay foundation for developing new theories and new methods in MSCI data pre-processing and application, and also support MSCI system manufacture, practical application, etc.

近年来，微波凝视关联成像技术被提出并逐渐成为国内外研究热点，其基于光学关联成像思想，通过时空两维随机辐射场成功实现了基于静止平台的高时空分辨率成像。然而现有研究主要集中在系统研制、关联重构等技术攻关工作，缺乏该新型成像体制下的应用基础理论研究。本项目以微波凝视关联成像体制下的分布目标散射特性研究为出发点，针对时空两维随机辐射源天线间耦合以及分布目标色散效应问题，通过天线耦合度分析、色散媒介时域有限差分法等途径，构建适于分布目标的散射特性模型和图像仿真方法；针对SAR图像特征模型难以适用关联图像特征描述问题，以典型线状类目标为例，构建基于自适应字典学习的目标稀疏表征方法并发展相应的目标检测识别模型；基于辐射场灵活可调以及多分辨率图像重构的特点，提出图像重构与检测识别一体化新方法。本项目研究可在微波凝视关联成像数据预处理与应用新理论、新方法方面奠定基础，为成像系统研制、实用化应用提供理论支撑。

微波凝视关联成像技术近些年来以其突破传统辐射场相干性、以阵列天线等实现随机辐射场并通过关联回波而成像的特点得到了快速的发展。此前的研究主要集中在系统研制、关联重构等技术攻关工作，缺乏该新型成像体制下的图像数据处理方法、基础理论的研究。本项目充分考虑微波凝视关联成像“时间换空间”成像特点，结合稀疏表示、深度学习等前沿数据处理理论，突破了微波凝视关联成像体制下的目标散射特征表达、压缩感知/深度学习理论与图像关联重构/目标检测识别过程的有机结合、辐射场灵活可调及图像多分辨率重构特点的有效应用等关键技术，重要结果包括：1）在微波凝视关联成像体制下的目标散射特性研究方面，系统推导分析了微波凝视关联成像机理和成像模型，对比分析了常规关联重建算法效果特别是噪声情况下重建效果，提出了基于稀疏学习字典的微波凝视关联图像重建方法，研制了一套微波凝视关联成像仿真分析系统，对目标散射特性进行了系统分析。2）在目标稀疏表达与检测识别模型研究方面，开展了微波图像目标特征稀疏表示方法研究，提出了面向目标识别的稀疏学习字典构建方法；在此基础上，提出了基于Gabor多尺度特征和稀疏表示的目标识别方法、基于Faster R-CNN的微波图像目标检测方法、基于级联特征CNN模型的微波图像目标识别算法，实验结果表明识别结果具有较高的鲁棒性和精度。3）在图像关联重构与检测识别一体化方法研究方面，利用辐射场灵活可调和可重构不同分辨率图像的特点，提出了针对背景均匀目标的关联成像策略、动态分辨率成像识别方法以及基于差分信息的运动目标快速探测模式，实现了微波凝视关联成像特性优势的利用。上述成果推动了微波凝视关联成像技术应用和微波图像目标检测识别理论的发展。