Edge-aware adaptive hyperspectral acquisition via synergistic dual-camera architecture

Snapshot hyperspectral imagers enable the acquisition of rich spatial and spectral information in a single exposure, making them a powerful tool for a wide range of high-speed applications. A prominent class of snapshot hyperspectral imaging methods reconstructs the full data cube from a high-resolution guide image combined with a set of spatially sparse spectral measurements. However, these methods frequently exhibit reconstruction artifacts and quality degradation primarily attributed to insufficient spectral sampling. To overcome this limitation, we propose an adaptive hyperspectral acquisition system built upon a synergistic edge-aware spectral sampling strategy and a corresponding reconstruction network. By considering the structure of the scene, our sampling strategy adaptively allocates measurement resources to better sense the scene, enabling enhanced scene characterization. The reconstruction network performs spatial-spectral dual filtering, employing 3D convolutional kernels within its spectral filters to effectively leverage joint spatial-spectral correlations for high-fidelity restoration. Comprehensive simulations and real-world experiments demonstrate that our integrated approach of adaptive sampling and deep reconstruction provides a robust and efficient solution for snapshot hyperspectral imaging.