Illumination color-coded and neural network decoding for snapshot hyperspectral imaging

Hyperspectral imaging (HSI) provides richer information, leading to its pivotal role in various applications. In this study, we propose an illumination color-coded and neural-decoding HSI method (INHSI) for physically reliable neural-based HSI. We leverage the inherent color-encoding capability of digital cameras for data capturing and employ physics-informed neural networks to reconstruct hyperspectral cubes. By systematically incorporating random illumination conditions into both the dataset and the reconstruction model, we effectively refine the spectral recovery process, achieving exceptional fidelity of direct spectral output. Our method is proved to be not only fast and cost-effective but also reliable, achieving high-quality performance in the experiment.