Single-photon CASSI: towards ultralow-light spectral imaging

Capturing spectral information in ultralow-light conditions remains challenging. Here, we propose a single-photon coded aperture snapshot spectral imaging (CASSI) system that incorporates the unique properties of single-photon avalanche diodes (SPAD) with the CASSI framework, efficiently integrating high sensitivity and spectral information. To tackle the challenges of sparse spatial sampling in directly integrated CASSI-SPAD systems and the lack of low-light spectral data and specialized networks, we developed an efficient spectral-spatial model for ultralow-light conditions, with an SPAD-CASSI simulation algorithm utilizing calibrated average photon counts, followed by noise and spectral-spatial encoding. Employing an exposure time of 100 µs at 0.15 lux, we simulated SPAD-CASSI data from 450 to 650 nm in 8 wavelength channels at 64 × 32 pixel resolution and designed a deep learning network for spectral reconstruction. Experimental results on macroscopic scenes, microscopic imaging, and high-speed scenarios show that our approach significantly improves the performance of snapshot spectral imaging in ultralow-light conditions, providing a robust solution for challenging imaging scenarios.