Spectral response characteristic-driven ICMOS color night vision imaging system and color reconstruction network

A spectral response characteristic-driven, intensified complementary metal-oxide semiconductor (ICMOS), color night vision imaging system and an associated color reconstruction network are proposed to address the challenges of nonlinear color shift and high intensity noise coupling in imaging under extremely low-light conditions in color night vision devices. An ICMOS-based low-light color imaging system was established to perform time-division low-light color imaging experiments and image acquisition based on spectral matching theory. To overcome the scarcity of high-quality, paired, and low-light color datasets, a physics-based synthetic data generation method is proposed. Integrating real-world multispectral data, environmental light source spectral data, and the system’s spectral response characteristics, a “low-light color cast–ground truth” dataset was constructed. Furthermore, a dual-branch color reconstruction network is designed. The proposed method effectively corrects severe color deviations caused by spectral response mismatch and improves the signal-to-noise ratio and color fidelity of images, achieving high-quality, natural-looking color night vision imaging under extreme low-light illuminance Levels of 4–5 (10⁻³–10⁻⁴ lx).