Low‐Cost High‐Performance VIS‐NIR Snapshot Imager via Single‐Exposure Patterning and Cumulative Attention Transformer Reconstruction

Abstract Visible and near‐infrared (VIS‐NIR) spectral imaging is vital for agriculture, food safety, and biomedical applications. Conventional spectral imaging relies on precision optical components with limited environmental adaptability, whereas computational spectral imaging employs advanced processing algorithms to simplify hardware architecture while improving system flexibility. However, developing compact, high‐performance, low‐cost snapshot systems remains challenging, especially in mask fabrication and real‐time imaging algorithms. In this work, a low‐cost snapshot VIS‐NIR spectral imager based on an on‐chip all‐dielectric weak‐confined Fabry–Pérot filter array and a deep learning‐based reconstruction approach is presented. Using single‐exposure patterning and the Cumulative Attention Transformer with Random Mask (CATRM) algorithm, the manufacturing process is streamlined while the reconstruction accuracy is enhanced. The system achieves high spatial resolution (100.17 lp mm −1 ) and maintains isotropic imaging fidelity, while reconstructing full‐field (2048 × 2048 × 61) hyperspectral data at 12.35 fps. The spectral accuracy of the imager is confirmed by spectral imaging of two traditional Chinese medicinal herbs, Astragalus membranaceus and Coptis chinensis , which shows over 99.1% cosine similarity between the system and a commercial scanning hyperspectral imager. Moreover, the broad application prospects are validated by non‐destructive sugar content prediction in mangoes. The integrated imager design enables compact, cost‐effective VIS‐NIR spectral imaging for diverse application scenarios.