A Detection Algorithm for Main Components of Highway Bridges Based on Improved YOLOv8n

Addressing the issues of low efficiency, high difficulty, and danger associated with traditional manual inspection of highway bridges, an improved detection algorithm for main components of highway bridges based on the YOLOv8n was proposed. Firstly, the MS-Block module was added into the backbone network of YOLOv8n, addressing the problems of missed and false detections caused by size differences among multiple targets. Subsequently, the CBAM attention mechanism was added, combining channel and spatial attention to capture key information in features and improve detection accuracy. The CIoU loss function of YOLOv8n was replaced with FocalCIoU to increase the weight of hard-to-classify samples and reduce classification errors. In addition, 1,200 images of highway bridges were collected, and LabelImg was used to annotate girders, bridge piers, and bank piers. Data augmentation was then applied to expand the training samples to three times their original number, creating a dataset of main components of highway bridges. Experiment results show that the improved model achieves mAP50, accuracy, and recall rates of 89.9%, 87.9%, and 85.4% respectively, representing improvements of 2.4%, 2.8%, and 2.3% compared to the original model. The model's parameters and computational volume increased by only 0.8M and 0.6G, respectively. This study effectively enhances detection performance and indicates that the improved model can be effectively applied to the task of object detection for main components of highway bridges.