Multiple Spatial‐Spectral Encoding Empowered High Fidelity Snapshot Spectral Imaging

Abstract Computational snapshot spectral imaging (C‐SSI) technique combining ingenious hardware encoding and powerful software decoding module has established a new paradigm for hyperspectral imaging. However, current encoding mechanisms still face limitations: the spatial‐spectral encoding model merges spatial and spectral information into a 2D measurement, making it difficult to decouple 3D datacube under high compression ratio; Multiple spectral encoding model requires excessive measurements, limiting imaging system design flexibility. To address these challenges, a Multiple SpAtial‐SpEctral encoding model (MSA‐SE) is proposed to implement efficient pixel‐level encoding with spatial multiplexing in this study. A corresponding masks fusion attention network (MFANet) is developed to reconstruct broadband spectra. Specially, spatial‐spectral synergistic correlation and information entropy metrics are introduced for the first time to comprehensively evaluate the encoding capacity of encoding element and analyze its underlying impact mechanisms on spectral reconstruction. Furthermore, to experimentally demonstrate the effectiveness of MSA‐SE encoding model, a C‐SSI system (CI‐ORRIS) is constructed based on the simple combination of a multi‐aperture lens array and an on‐chip continuous variable filter (CVF). The system produces spectral images of 66 channels during 440–700 nm, achieving a maximum spectral resolution of 3.7 nm, which is nearly 5 times higher than that of the original resolution of CVF.