Research on a dual-camera coded aperture snapshot spectral imaging system

This study focuses on a dual-camera coded aperture snapshot spectral imaging system. The optical system was designed to cover the spectral range of 450–670 nm with a spatial resolution of 2048 × 2048 and 118 spectral channels. Based on the design, key optical components were fabricated and procured, and a prototype system was constructed. To enhance reconstruction quality, we propose an RGB-guided covariance-driven sparse reconstruction method within a Plug-and-Play framework, termed GCSR-PnP (guided covariance sparse representation—plug-and-play). This method leverages non-local structural priors extracted from the RGB modality to guide hyperspectral recovery. To enable effective fusion of dual-channel data, a coarse-to-fine pixel-level registration strategy based on normalized cross-correlation (NCC) is designed, achieving robust alignment between RGB and CASSI images under significant cross-modal differences. On this basis, GCSR-PnP is evaluated on subsets of the CAVE and KAIST datasets and compared with current mainstream hyperspectral reconstruction methods, including PnP-FFDNet (pretrained deep prior) and PnP-DIP (unsupervised adaptive prior). Finally, real-scene experiments are conducted using the developed imaging system and GCSR-PnP model. Results demonstrate that the proposed method achieves superior spatial detail preservation and high spectral fidelity. The proposed system and method strongly support the engineering application and high-quality reconstruction of compressive spectral imaging technologies.