基于深度学习的太赫兹旋转稀疏阵列雷达成像

With the promulgation and implementation of the national anti-terrorism act, the security detection of human and belongings in the public places becomes an essential means for preventing public security events. Terahertz (THz) wave is located between the millimeter and infrared regions in the electromagnetic spectrum, which brings the THz imaging with many advantages, such as high spatial resolution, penetration capability, and safety characteristics. Therefore, THz imaging is a new reliable security technique. To overcome the difficulty that existing security systems faced in the real-time imaging of moving human, the fan-shaped rotating sparse array imaging for human is proposed by this project. The proposed imaging scheme combines real arrays and mechanical scanning together, and unites real aperture and synthetic aperture together, which contributes to faster scanning speed, less antennas and controllable cost. Complex-valued convolutional neural network, as one of deep learning methods, will be proposed to real the automatic feature-enhanced imaging. Thus, the real-time high resolution imaging of moving human can be achieved, and the efficiency of security inspection can be improved significantly, therefore showing wide application prospects in the future.

随着《反恐怖主义法》的颁布实施，在公共安全场所对人员及物品进行安全检查成为预防公共安全事件的必要手段。太赫兹波处在毫米波向红外可见光过渡的波段(0.1-10THz)，兼具微波和红外的优势，具有成像分辨率高、穿透能力强、对人体无伤害等特点，因此，太赫兹成像是一项新的可靠安检技术。本项目以突破现有毫米波/太赫兹安检系统在运动人体实时成像方面的难题为出发点，利用实阵列与机械扫描结合、实孔径与虚拟孔径结合，提出风扇式竖直面旋转的稀疏线阵太赫兹雷达人体成像体制，具有旋转扫描速度快、阵元数目少、成本可控等优点，为自动增强可疑物品特征，提出并研究基于复数卷积神经网络深度学习的成像方法，可实现对无驻留人体的高精度实时成像，有望大幅提升安检效果和效率，在太赫兹雷达成像、安检探测领域有着广阔应用前景。

为实现近场条件下对可疑目标探测识别，本项目融合合成孔径和线性阵列思想，提出并实现了竖直平面旋转式稀疏阵列太赫兹雷达成像技术。利用快速风扇式扫描形成的圆面孔径获得二维横向高分辨，利用太赫兹大带宽获得距离向高分辨，避免了传统平面扫描方式的频繁加减速过程，具有旋转扫描速度快等优点，同时仅通过小规模硬件阵列就可实现大规模阵列效果。基于该阵列体制，对旋转稀疏阵列设计与回波建模、图像学习样本获取方法、深度学习稀疏阵列成像方法等问题展开了深入研究，实现了对近场目标的高精度近实时成像。项目成果有望在要地防护等实际应用场景中发挥重要作用，能够大幅提升成像识别效率，在非合作、无感化、大流量安检环境下具有广阔的应用前景，同时对雷达成像体制与技术发展变革起到一定的推动作用。