Daytime star spot detection based on spatial-temporal feature context enhancement for optical star sensors

Optical star sensors provide high-precision attitude determination for astronomical navigation, and daytime star sensors further enable all-time operation by breaking through traditional daytime constraints. However, intense daytime atmospheric radiation, along with optical lens and detector imperfections, reduces the signal-to-noise ratio (SNR) of dim star spots and causes numerous false alarms, which cannot be effectively addressed by traditional spatial feature-based methods. To tackle this problem, this paper proposes a multi-stream feature fusion framework integrated with a frame difference-optical flow guided spatial-temporal enhancement module, which supplements motion-aware features and adopts attention mechanisms for noise suppression. Validated via synthetic dataset training and ground-based experiments on real star maps, the proposed approach achieves superior daytime star detection performance, thus enhancing the robustness of star sensors in harsh optical environments.