An Innovative Approach for Detecting Bridge Defects Based on UAV Imagery in Low-Light Environments

Bridge inspection plays a pivotal role in maintaining bridge safety and structural integrity, which is a critical task in the civil engineering industry. The emerging adoption of UAV (Unmanned Aerial Vehicle) image-based inspection, coupled with AI-driven defect detection models, offers a swift, secure, and cost-effective solution for maintaining bridges. However, when inspecting low-light environments, such as areas under bridges, the poor illumination and increased noise in the image often lead to issues with false detections or missed detection. Regarding the problem, this paper applies image enhancement techniques to low-light under-bridge images and evaluates the performance of commonly used deep learning-based detectors for defect detection. To improve detection performance, an attention mechanism CBAM is incorporated into YOLOv5. Results demonstrate that the proposed CBAM-YOLOv5 algorithm can improve detection accuracy by 2.2%-8.1% compared to other object detectors. © 2023 30th EG-ICE: International Conference on Intelligent Computing in Engineering 2023. All rights reserved.