Reconstructing visible-near-infrared hyperspectral images via liquid crystal based optoelectronic hybrid neural network

作     者：Li, Shiqi Li, Hui Li, Tian Song, Xiaoyue Wu, Yuntao 

作者机构：Wuhan Inst Technol Sch Comp Sci & Engn Wuhan 430205 Peoples R China Hubei Key Lab Intelligent Robot Wuhan 430205 Peoples R China 

出 版 物：《OPTICS AND LASER TECHNOLOGY》 (Opt Laser Technol)

年 卷 期：2025年第184卷

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：National Natural Science Foundation of China Innovative Group Program of Natural Science Foundation of Hubei Province [2023AFA035] Hubei Provincial Key Laboratory of Intelligent Robot [HBIRL202203] Engineering Research Center of Digital Imaging and Display, Ministry of Education, Soochow University [SDGC2134] Shanxi Key Laboratory of Singal Capturing & Processing, North University of China [2023-001] Graduate Innovative Fund of Wuhan Institute of Technology [CX2023298, CX2023299] Knowledge Innovation Program of Wuhan-Basic Research 61771353 

主　　题：Visible-near-infrared hyperspectral reconstruction Optoelectronic Hybrid Neural Network Digital prism Liquid crystal microlens array 

摘      要：We introduce an Optoelectronic Hybrid Neural Network (OHNN) that integrates software and hardware optimizations to enable the simultaneous reconstruction of visible-near-infrared (VIS-NIR) hyperspectral images. The network employs a linear polarizer (LP) and liquid crystal (LC) microlens array (MLA). We harness the polarization and electrically controlled tunable properties of LC-MLA in conjunction with digital optics technology to construct a software-defined dispersion element digital prism (DP). The OHNN can reconstruct VIS-NIR hyperspectral images, ranging from 400 nm to 900 nm, from limited datasets with a spectral accuracy of 10 nm. The reconstructed hyperspectral images boast a Peak signal-to-noise ratio of approximately 31 dB, marking a 3.2 % improvement over existing state-of-the-art methods.