Abstract Concrete bridges are critical components of urban infrastructure, and their structural health directly influences the safety and efficiency of urban transportation. However, existing bridge defect detection methods often focus on single defect types and require manual data collection, which is both time-consuming and labor-intensive. Although recent advancements in UAV technology have significantly improved the efficiency of image acquisition, challenges such as varying viewing angles, illumination conditions, and complex environmental backgrounds in captured images continue to hinder the accuracy of existing methods for bridge defect detection. To address these limitations, we propose CDC-YOLO, an improved defect detection network designed for multi-type defect detection in concrete bridges based on UAV images. In this proposed CDC-YOLO model, a multi-dimensional feature extraction module is presented to capture both shallow and detailed features, ensuring the accurate identification of fine crack defects in real-world scenarios. Then, a dynamic feature recombination module is proposed to improve the adaptability of the network in detecting irregular defect shapes and distributions in complex scenes. In addition, an adaptive feature fusion module is designed, which overcomes the limitations of traditional fusion methods by effectively mitigating false positives and missed detections caused by complex environmental interference, illumination changes, and varying viewing angles. Finally, various experiments are conducted, and the quantitative and qualitative results demonstrate the superior performance of the proposed model over state-of-the-art defect detection methods, particularly in detecting small defects with low contrast against their environmental backgrounds.
Ni et al. (2025) studied this question.