DP-PSSI: Data-Driven Prior for Pharmaceutical Snapshot Spectral Imaging

Hyperspectral images contain rich spatial and spectral information, leading to wide applications in many scenarios. Especially in the pharmaceutical field, high spatial and spectral resolution imaging of Active Pharmaceutical Ingredient (API) has an important impact on drug quality detection and analysis. However, existing snapshot computational imaging methods are rarely used in private scenarios, and it is difficult for existing methods to effectively improve the spatial and spectral resolution of API. Therefore, this paper proposes a data-driven pharmaceutical hyperspectral image prior based on Coded Aperture Snapshot Spectral Imager (CASSI) technology. This prior integrates the characteristics of low rank and non-local self-similarity in hyperspectral depth semantic information. And we propose a dense block network in the spatial feature extraction module of Unet which greatly strengthens the feature propagation in solving the problem of extracting detailed texture features from the spatial information of API. At the same time, this method improves the spectral fidelity of the reconstructed data and is conducive to improving the spectral analysis capabilities of API. Our proposed method improves the Peak Signal-to-Noise Ratio (PSNR) by 5.57% and the Structural Similarity Index Measure (SSIM) by 10.99% compared with SOTA method. Furthermore, we designed a prototype of the pharmaceutical snapshot spectral imager to facilitate the experiments, and our API imaging experiments on COVID-19 therapeutic drugs demonstrated the superiority of our method.