Compressed hyperspectral imaging based on residual-spectral attention mechanism and similar image prior

• Published in: Optics and lasers in engineering Vol. 180; p. 108330
• Main Authors: Qu, Xiaorui, Zhao, Jufeng, Cheng, Yanhui, Tian, Haijun, Cui, Guangmang
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2024
• Subjects: Compressive spectral imaging; Residual spectral attention prior; Spectral reflectance; Residual spectral attention prior; Compressive spectral imaging; Spectral reflectance
• ISSN: 0143-8166; 1873-0302
• DOI: 10.1016/j.optlaseng.2024.108330

Hyperspectral imaging based on compression coding is one of the mainstream imaging methods. The methods relying on deep image priors currently fail to fully investigate the characteristics of spectral data and imaging. This constrains the enhancement of both the efficiency and quality of the reconstruction process. We propose a hyperspectral imaging reconstruction framework based on Spectral Reflactance-Residual Spectral attention Prior (SR-RSP). Specifically, we propose a Spectral Reflectance (SR) algorithm to quickly generate effective spectral data to innovate the deep image prior. In this way, the network reconstruction quality can be improved. In addition, considering the uncertainty of the reconstruction process and the multi-channel characteristics of the spectrum, a Residual Spectral attention Prior is proposed. We combine it with SR to form SR-RSP. By improving the fitting ability of the network, the spectral reconstruction quality and spatial resolution are improved. Our method has significantly improved spectral reconstruction quality and spatial resolution compared with traditional state-of-the-art methods. Simulation results and actual data show that compared with similar advanced methods, our method can significantly improve the quality of spectral reconstruction and further reduce the time cost of reconstruction.