太赫兹波相对于光波和红外而言, 波长较长, 因此太赫兹成像系统的空间分辨率较低。为了在一定程度上提高太赫兹图像的分辨率, 提出了一种超分辨率重建方法。该方法基于深度学习, 在原有残差结构上进行改进, 加深了网络结构。对太赫兹图像进行训练, 拟合低分辨率和高分辨率图像之间的映射关系, 能够实现在单个网络中同时处理多个尺度的超分辨率图像。实验结果表明, 相对于一些传统的超分辨率重建方法, 本方法在量化指标和视觉效果上都有一定程度的提升, 且处理速度相对较快, 能够满足安检系统中实时处理的要求。

Terahertz waves have longer wavelengths compared with light waves and infrared, so the spatial resolution of terahertz imaging systems is lower. Herein, a super-resolution reconstruction method has been proposed to improve the resolution of terahertz images. The method deepens the network structure due to the improvements on the original residual structure based on deep learning. Furthermore, training terahertz images to fit the mapping between low-resolution and high-resolution images enables simultaneous processing of multiple-scale super-resolution images in a single network. The experimental results indicate that the method has a certain degree of improvement in quantitative indicators and visual effects, and the processing speed is relatively fast, which can meet the requirements of real-time processing in the security inspection system compared with some traditional super-resolution reconstruction methods.