Abstract: Spacecraft operating in complex space environments are susceptible to micro-scale surface damage, which poses threats to structural integrity and mission reliability. To overcome the limitations in scale modeling, feature alignment, and environmental adaptability of existing detection techniques, a novel spacecraft surface micro-damage detection method based on a dual-branch multi-scale Vision Transformer (ViT) is proposed. In the proposed architecture, global environmental context and local high-resolution detail features are extracted, and a saliency-guided fusion module (SGFM) is employed to dynamically aggregate critical regions and enhance the detection accuracy of minute damages. The method was evaluated on the Spacecraft-DS dataset for impurity and scratch detection on metallic housings within an end-to-end training framework. Compared with mainstream models such as YOLO-V8 and DETR, the proposed approach achieved improvements of up to 2.2% in APall, Recall, and F1, and reached an inference speed of 25 FPS. Superior performance was observed particularly in small-sized damage recognition and complex-background suppression, effectively addressing the limited receptive field of traditional CNNs and the scale-adaptability issues of standard ViTs. The results demonstrate that the proposed method provides an efficient technical solution for applications such as space-station inspection and satellite degradation analysis.