连续波太赫兹成像系统的单幅图像超分辨重建  

Single-image super-resolution reconstruction for continuous-wave terahertz imaging systems

作　　者：王欢 郎利影[2,3] 庞亚军 张雷 郑伟 席思星 Wang Huan;Lang Liying;Pang Yajun;Zhang Lei;Zheng Wei;Xi Sixing(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China;Center for Advanced Laser Technology,Hebei University of Technology,Tianjin 300401,China;Hebei Key Laboratory of Advanced Laser Technology and Equipment,Tianjin 300401,China;School of Mathematics&Physics Science and Engineering,Hebei University of Engineering,Handan 075000,China;Department of Physics and Electronic Engineering,Yuncheng University,Yuncheng 044000,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300401 [2]河北工业大学先进激光技术研究中心,天津300401 [3]河北省先进激光技术与装备重点实验室,天津300401 [4]河北工程大学数理科学与工程学院,河北邯郸075000 [5]运城学院物理与电子工程系,山西运城044000

出　　处：《红外与激光工程》2023年第1期263-270,共8页Infrared and Laser Engineering

基　　金：国家自然科学基金(11904073);河北省重点研发项目(20371802D);河北省自然科学基金(F2019402351);河北省教育厅青年拔尖人才项目(BJ2020028)。

摘　　要：针对现有的太赫兹成像系统所需硬件设备复杂且昂贵的问题,设计了基于单幅图像超分辨重建的连续波太赫兹成像系统,降低设备复杂度和硬件成本。通过对该成像系统生成的太赫兹图像进行双维度预处理,降低图像处理的占用内存,提高后续处理速度。引入限制对比度自适应直方图均衡方法对太赫兹图像进行分区域对比度提升,有效解决太赫兹图像对比度低的问题。利用稀疏表示和字典学习实现太赫兹图像的超分辨重建,提出了反余割拟牛顿平滑零范数的算法解决零范数优化问题,提高了重建精度。通过对该成像系统采集的单幅太赫兹图像进行超分辨重建,在边缘强度上提高了3.232,在平均梯度对比中提高了0.300,验证了对单幅太赫兹图像超分辨重建的有效性与优越性。To address the problem that existing terahertz imaging systems require complex and expensive hardware equipment, a continuous-wave terahertz imaging system based on single-image super-resolution reconstruction is designed to reduce equipment complexity and hardware cost. By preprocessing the terahertz images generated by this imaging system in two dimensions, the occupied memory of image processing is reduced and the speed of subsequent processing is increased. A restricted-contrast adaptive histogram equalization algorithm is introduced for sub-regional contrast enhancement of terahertz images to effectively solve the problem of low contrast of terahertz images. The super-resolution reconstruction of terahertz images is achieved by using sparse representation and dictionary learning, and the algorithm of inverse cosecant fitted with Newtonian smoothing zero parity is proposed to solve the zero-norm optimization problem and improve the reconstruction accuracy. By performing super-resolution reconstruction of single terahertz images acquired by this imaging system, the algorithm improves 3.232 in edge intensity and 0.300 in mean gradient comparison, which verifies the effectiveness and superiority of super-resolution reconstruction of single terahertz images.

关 键 词：太赫兹成像 连续太赫兹波 超分辨重建 

分 类 号：TN219[电子电信—物理电子学] TP399[自动化与计算机技术—计算机应用技术]