基于全变分与深度去噪网络的压缩超快摄像重构方法  

Total variation and FastDVDNet based reconstruction method for compressed ultrafast photography

作　　者：龙佳乐[1] 李英荣 丁毅 马钊 张建民[1] 许彬 LONG Jiale;LI Yingrong;DING Yi;MA Zhao;ZHANG Jianmin;XU Bin(Faculty of Electronic and Information Engineering,Wuyi University,Jiangmen529020,China;Optics Valley Laboratory,Wuhan430074,China)

机构地区：[1]五邑大学电子与信息工程学院,广东江门529000 [2]光谷实验室,湖北武汉430000

出　　处：《光学技术》2025年第2期136-140,202,共6页Optical Technique

基　　金：广东省普通高校创新团队项目(2020KCXTD051);江门市基础与应用基础研究重点项目(2021030103730007331);江科[2023]111号;2023年度江门市基础与理论科学研究类科技计划项目(2023JC01002,2023JC01004);五邑大学港澳联合基金项目(2021WGALH17)。

摘　　要：利用压缩超快摄像技术,有希望以极高的时间分辨率揭示飞秒激光脉冲在介质中的瞬态散射过程。然而,基于两步迭代收缩/阈值算法(TwIST)的压缩超快摄像重构由于受到欠采样策略引起的超高数据压缩率影响,存在图像重构质量不高、结果不稳定等问题。为克服这些问题,本文提出了一种交替使用全变分(TV)与深度去噪网络(FastDVDNet)结合的即插即用广义交替投影框架重建算法,以改善压缩超快摄影的图像质量。实验结果表明,与传统的两步迭代收缩/阈值算法相比,文章提出的算法可以显著提高超快图像序列的重构质量。Using compressed ultrafast photography,it is promising to reveal the transient scattering process of femtosecond laser pulses in a medium with very high temporal resolution.However,based on the two-step iterative shrinkage/thresholding algorithm(Two-step Iterative Shrinkage/Thresholding(TwIST))based compressed ultrafast camera reconstruction suffers from poor image reconstruction quality and unstable results due to the ultra-high data compression rate caused by the under sampling strategy.To overcome these problems,a method is proposed in which the use of Total Variation(TV)combined with a deep denoising network(FastDVDNet)is alternated as a plug-and-play generalized alternating projection frame reconstruction algorithm to improve the image quality of compressed ultrafast photography.Experimental results show that the algorithm proposed in this paper can significantly improve the reconstruction quality of ultrafast image sequences compared to the traditional two-step iterative shrinkage/thresholding algorithm.

关 键 词：压缩超快摄影 图像重建 飞秒激光脉冲 重构算法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]