<p>Timely and effective detection of surface defects on bridges is crucial for ensuring public transportation safety and extending the lifespan of bridges. Traditional bridge defect detection primarily relies on manual inspections, which are highly subjective, inefficient, and exhibit a significant rate of undetected issues. Moreover, there is a lack of effective detection methods for bridges located in special areas, such as large-span canyon bridges. To address these challenges, this study focuses on the Chishi Bridge in Chenzhou, Hunan Province, China, which is the largest multi-tower concrete cable-stayed bridge in the world with high piers. First, we designed a fully automated detection system for one-stop intelligent maintenance operations. Second, our system integrates a bridge damage detection algorithm guided by fractal geometric features. This algorithm is an improvement based on the classic object detection algorithm YOLOv7 (You Only Look Once), incorporating the SimAM and CARAFE attention mechanisms to enhance damage feature recognition, accurately identifying and assessing various defects on the underside of the bridge. The proposed method achieves an average mean precision (mAP) of 87.24% and reduces inference time by 15%, demonstrating significant improvements in both accuracy and efficiency compared to the baseline. Overall, this significantly enhances the level of intelligent defect detection for the Chishi Bridge, ensuring safe and stable operation while showcasing considerable economic production value.</p>