• Proposes the first ground-based dataset of BeiDou navigation satellites specifically designed for on-orbit low-light image enhancement (LLIE), collected via a robotic simulation testbed that mimics space lighting conditions.
• Introduces a collision-free workspace and pose-stratified sampling strategy to ensure diverse and safe data collection across different satellite orientations and distances.
• Develops a novel diffusion model with fused attention guidance that enhances image contrast and reveals dark-region details without over-exposure or blurred output.
• Demonstrates superior on-orbit LLIE performance compared to prior methods, validating the effectiveness of the proposed dataset and enhancement framework.