Fourier single-pixel spectral imaging via local low-rank tensor nuclear norm and deep tensor priors

作     者：Tang, Zixin Hu, Yexun Duo, Chen Yang, Guowei Jiang, Taixiang Guo, Daqing 

作者机构：School of Computing and Artificial Intelligence Southwestern University of Finance and Economics Sichuan Province Chengdu China Kash Institute of Electronics and Information Industry Kash China Engineering Research Center of Intelligent Finance Ministry of Education Southwestern University of Finance and Economics China Chongqing University of Education School of Artificial Intelligence China School of Life Science and Technology University of Electronic Science and Technology of China Sichuan Province Chengdu China Chongqing University Industrial Technology Research Institute Department of Science and Technology Development China 

出 版 物：《Optics Letters》 (Opt. Lett.)

年 卷 期：2025年第50卷第4期

页      面：1281-1284页

核心收录：

学科分类：070207[理学-光学] 07[理学] 08[工学] 0803[工学-光学工程] 0702[理学-物理学] 

基　　金：Natural Science Foundation of Xinjiang Uygur Autonomous Region (2024D01B06, 2024D01A18) Sichuan Provincial Science and Technology Support Program (24NSFSC1452, 2024ZYD0147) Natural Science Foundation of Chongqing Municipality (CSTB2024NSCQMSX0627) Science and Technology Research Program of Chongqing Municipal Education Commission (KJZD-K202401603) National Key Research and Development Program of China (2023YFF1204200) The Graduate Representative Achievement Cultivation Project of SWUFE (JGS2024069) China Postdoctoral Science Foundation (2024M763876) 

主　　题：Tensors 

摘      要：The imaging quality of single-pixel spectral imaging (SSI) methods is poor at a low sampling ratio (SR). To tackle this problem, a new Fourier single-pixel spectral imaging (FSSI) technique is proposed. Firstly, we introduce the low-rank tensor nuclear norm (TNN) to characterize the correlation between spectral images. Compared with the conventional method, TNN reconstructs image details better but brings image artifacts simultaneously. Therefore, local low-rank TNN (LTNN) constraint is proposed to ameliorate global ones and to reduce the distortion caused by TNN and low SR. Secondly, to make full use of the spectral information, the proposed constraint is used as the coarse prior, and the deep tensor prior (DTP) is introduced as the fine one to construct the joint priors. Different from the single prior, the joint method can make the two priors benefit and improve each other and further enhance the imaging quality. Finally, an efficient and high-quality SSI technique is achieved by deducing the closed-form solution algorithm. Experimental results show that our method significantly improves the quality of FSSI as much as 7–10 dB when compared to 3DTV at the SR of 5%. © 2025 Optica Publishing Group. All rights, including for text and data mining (TDM), Artificial Intelligence (AI) training, and similar technologies, are reserved.