The application of deep learning in bridge surface disease detection

To improve the efficiency and accuracy of bridge damage target detection, this paper proposes an improved YOLOv8 detection algorithm, KAC-YOLO, which integrates the KAN network, Attention-based Intra-Feature Interaction in Scale (AIFI), and the SIoU loss function. The study enhances feature extraction through the KAN network and improves sensitivity to multi-scale damage features by utilizing the AIFI mechanism for cross-scale feature interaction. Additionally, the SIoU loss function is applied to accelerate model convergence. From a system design perspective, the architecture and modules were optimized to ensure detection efficiency and accuracy. Experimental results show that the improved YOLOv8 model significantly boosts average accuracy and detection speed on the bridge damage dataset. Compared with the original model, KAC-YOLO not only improves detection performance but also demonstrates better adaptability and robustness, providing an efficient and precise solution for bridge structural health monitoring.