关键词: 裂缝检测, YOLOv5, 注意力机制, 自适应空间特征融合

Abstract: The emergence of road cracks has an obvious impact on the road service life and the safety of people and vehicles,so it is necessary to detect road cracks in time.Aiming at the problems of low detection accuracy caused by small crack target and complex image background in UAV image,this paper takes the crack image collected by UAV as research data,and proposes a deep learning road crack detection method based on improved YOLOv5 model.The attention mechanisms of CBAM,SimAM and CA are added to the backbone network of YOLOv5 model to improve the crack recognition ability and detection accuracy of the model.Comparative analysis is carried out through ablation experiments.At the same time,adaptive spatial feature fusion algorithm is incorporated into the YOLOv5 model to improve the ability of crack feature extraction.The research shows that the accuracy of the improved YOLOv5 network model is significantly higher than that of the original model,and the mean average accuracy(mAP) is increased by 20.6%.It not only ensures the accuracy but also effectively improves the detection accuracy,and can provide a new method for road crack detection.

Key words: crack detection, YOLOv5, attention mechanism, adaptive spatial feature fusion