In recent years,the bridge construction in China has developed unprecedentedly,and the problem of bridge safety operation and maintenance has become more and more prominent.In this paper,we propose a crack detection model YOLOv8-BC based on YOLOv8,which adopts GhostConv's improved structure of HGNetv2 to replace the backbone net-work of YOLOv8,to realize the model's lightweight;introduces the deformable attention mechanism DAttention,to obtain more spatial information;and uses the auxiliary edge-based inner-iou loss function based on auxiliary edges instead of CIOU loss function to improve the generalization ability and detec-tion accuracy of the model.Experiments on the self-con-structed dataset show that the YOLOv8-BC model achieves an accuracy of 95.3％and the mAP50 reaches 95.9％,which is improved by 2.3％and 2.0％,respectively,compared with the YOLOv8s model,while the number of parameters is reduced by 15.3％.This indicates that the YOLOv8-BC model can detect bridge cracks in complex environments more efficiently,which provides important ideas and technical support for in-dustrial inspection.
维普
万方数据
