Hybrid mosaic-array design for on-chip computational hyperspectral imaging: balancing accuracy and adaptability

Hyperspectral imaging simultaneously captures spatial and spectral data, offering broad applicability across numerous fields. Traditional spectral imaging systems based on dispersive elements or narrow-band filters are limited by bulky designs, low light throughput, slow acquisition times, and an inherent trade-off between spatial and spectral resolution. Computational spectral imaging addresses these challenges by encoding and sampling incident light and employing reconstruction algorithms to recover spectral information. Although on-chip mosaic-array approaches have recently attracted attention, most existing methods prioritize reconstruction accuracy while neglecting adaptability to varying imaging conditions. In this paper, we propose a novel 3×3 mosaic filter array that incorporates three random filters to enhance reconstruction accuracy and six narrow-band filters to improve adaptability under different illumination conditions. Compared to conventional methods that use solely narrow-band or random filters, our hybrid design achieves superior performance in both accuracy and generalization. This innovative solution effectively balances spectral reconstruction precision with scene adaptability, advancing the capabilities of on-chip computational hyperspectral imaging.