可重构自适应毫米波快速安检成像与自动目标识别关键技术研究

In recent years, domestic and international anti-terrorism situation is becoming increasingly serious. The pressure on human security on some important occasions is huge. Research on millimeter-wave security imaging instrument and relevant core technology is of great theoretical and social significance.. The project plans to develop the new generation millimeter-wave security imaging theory and experimental research on long-distance and non-cooperation target body. Based on the theoretical deduction, large-scale three-dimensional electromagnetic simulation and construction of experimental system, we deeply investigate key scientific problems on the scattering theory of the micro-motion model of the human body under walking state, the optimal array distribution theory of the millimeter-wave space-frequency multi-dimensional reconfigurable sparse array, the structure and physical parameters’ relationship of neural network models for automatic target recognition applied to moving target. We strive to make breakthroughs on key technologies such as the long-distance and high-resolution three-dimensional reconfigurable sparse imaging, the compensation holographic imaging algorithm for moving human body based on the human body micro-motion model, the automatic target recognition algorithm for the human body with dangerous goods under walking state.. The project will develop a millimeter-wave non-cooperation target security imaging prototype which can image pedestrian target within distance of 1-10 meters, imaging time less than 1 second, and imaging resolution less than 5 cm (when target distance is 10 meters). A series of imaging algorithms are validated by the experimental prototype, which establishes a solid foundation for the practical application of long-distance and non-cooperative millimeter-wave human security imaging system.

近年来，国内外反恐形势日益严峻，重要场合的人体安检压力巨大，研究毫米波安检成像仪核心技术具有极其重要理论和社会意义。本项目拟开展新一代远距离非合作目标人物的毫米波安检成像理论与实验研究。通过理论推演、大规模三维电磁仿真以及搭建实验系统，对行走状态下的人体微动模型电磁散射机理、毫米波空间-频率多维可重构稀疏阵列的最优布阵理论、应用于运动目标的自动目标识别神经网络模型的结构以及物理参量关系等关键科学问题进行深入研究；在远距离高分辨率三维可重构稀疏成像、基于人体微动模型的运动人体补偿全息成像算法、行走状态下人体携带危险品自动目标识别算法等关键技术上取得突破。本项目将研制毫米波非合作目标安检成像实验样机一台，可对1-10米距离的行人目标进行成像，成像时间小于1秒，成像分辨率小于5厘米（目标距离为10米时）。通过实验样机验证一系列成像算法，为远距离非合作毫米波人体安检成像系统的实用化奠定坚实的基础。

近年来，国内外反恐形势日益严峻，针对行人的不停留人体安检需求日益增多，因此开展新一代远距离非合作目标人物的毫米波安检成像理论与实验研究具有极其重要理论和社会意义。针对行人走动成像需要解决的大数据量和运动成像问题，项目组提出了单发多收并行处理的毫米波稀疏阵面系统结构兼顾数据采集实时性和系统成本；提出基于时域后向散射、随机稀疏阵元以及随机稀疏频点的稀疏阵面成像算法以在保证成像质量前提下大幅度降低数据采集量；提出采用深度相机和卡尔曼滤波的人体运动状态参数提取算法、分区域抑制邻近通道的毫米波成像算法来提升运动人体目标的成像质量；提出多种毫米波人体违禁品自动目标识别算法来提高违禁品自动目标识别性能，最终完成了针对行人走动的非合作目标毫米波成像安检样机，并将项目研究的各种算法部署到设备中。项目研制的实验样机最远成像距离达到10m，并且在10m距离处成像分辨率达到5cm，可对行人目标进行成像，成像时间小于1秒，性能指标满足计划书要求。通过本项目的研究，为远距离非合作毫米波人体安检成像系统的实用化奠定坚实的基础。