面向工业应用的主动电子太赫兹成像：从硬件到人工智能驱动的范式转变
Active Electronic Terahertz Imaging for Industrial Applications: From Hardware to the Paradigm Shift by Artificial Intelligence
浏览论文内容
中文总结 AI 辅助
本综述聚焦主动电子太赫兹成像，综述其硬件、成像模态、增强技术及AI应用，推导技术路线图以解决太赫兹成像瓶颈，有望为工业应用提供高性价比解决方案。
中文摘要 AI 辅助
太赫兹（THz，0.3-10 THz）辐射成像具备独特的组合优势：可穿透干燥、非极性包装材料；介电函数变化可提供成像对比度；部分材料存在光谱指纹共振；光子能量无电离作用，对人类安全；从低频侧看，其拓展了微波雷达的能力至更高频率，从而获得显著更优的空间分辨率，且其波长仍可直接测量复值辐射场。本综述聚焦于结合电子源与功率探测器或相干接收机的主动太赫兹成像——这类系统最有望为质量控制、无损检测、安全筛查及态势感知测距等各类工业应用提供快速（理想情况下为实时）、高性价比且可部署的解决方案，且可部署于机器人及新兴自主平台。我们综述了紧凑型半导体探测器阵列的最新进展、成像模态（涵盖聚焦光束光栅、调频连续波架构至相干傅里叶平面采集），以及压缩感知等增强技术。特别关注人工智能日益增长的作用：从用于相位 Retrieval（相位恢复）与图像重建的卷积神经网络、物理信息深度学习，到用于自主系统设计的智能体框架。结合这些创新重新审视太赫兹成像的瓶颈——采集速度、分辨率、对比度与成本，并推导了以参考为基准的技术路线图，该路线图预计太赫兹成像的测量需求将降低一个数量级。
英文摘要
Imaging with terahertz (THz) radiation (0.3-10 THz) benefits from a unique combination of attributes: penetration through dry, non-polar packaging materials; variations of dielectric functions to provide contrast; the existence of spectral fingerprint resonances for some classes of materials; non-ionizing photon energies that are safe for use around humans; and - viewed from the low-frequency side - an extension of the capabilities of microwave radar to higher frequencies and thus to substantially better spatial resolution, at wavelengths which still permit direct measurement of the complex-valued radiation field. This review concentrates on active THz imaging with electronic sources combined with power detectors or coherent receivers - the system class most likely to deliver fast (ideally real-time), cost-effective and deployable solutions for a wide range of industrial applications such as quality control, non-destructive testing, security screening and ranging for situational awareness. Such systems should be deployable on robotic and emerging autonomous platforms. We review the state of the art of compact semiconductor detector arrays, of imaging modalities ranging from focused-beam raster and frequency-modulated continuous-wave architectures to coherent Fourier-plane acquisition, and of augmentation techniques such as compressive sensing. Particular attention is paid to the growing role of artificial intelligence: from convolutional neural networks and physics-informed deep learning for phase retrieval and image reconstruction, to agentic frameworks for autonomous system design. The bottlenecks of THz imaging - acquisition speed, resolution, contrast and cost - are re-examined in the light of these innovations, and a reference-anchored technology roadmap is derived which projects an order-of-magnitude reduction in the measurement requirements of THz imaging.
