基于改进YOLOv5s算法的桥梁裂缝自动检测及分类

Automatic Detection and Classification of Bridge Cracks Based on Improved YOLOv5s Algorithm

为了克服传统人工裂缝检测方法费时费力、局限性大等弊端与不足,提出一种基于深度学习的裂缝自动检测与分类方法。采用YOLOv5s算法作为基础,引入2种不同的注意力机制——SENet和Coordinate Attention,这些机制从大量数据中快速筛选出高价值信息,从而提高了YOLOv5s模型在裂缝识别和分类方面的效率。原始的YOLOv5s模型在1500张包含4种类型裂缝的图像上的检测结果为89.2%,引入注意力机制后,模型精度分别提高了5.7%和7.1%,达到了94.9%和96.3%。结果表明,改进后的YOLOv5s算法可以实现桥梁裂缝的自动检测及分类,在实际的桥梁性能测试中具有广泛应用前景。

A deep learning-based automatic crack detection and classification method is proposed to over come the shortcomings and limitations of traditional manual crack detection methods,which are time-consuming,labor-intensive,and of limited use.The YOLOv5s algorithm is used as the foundation,and two different attention mechanisms,SENet and Coordinate Attention,are introduced.Within a large amount of data,the high-value information is quickly filtered out by these mechanisms,thereby improving the efficiency of the YOLOv5s model in crack detection and classification.The detection accuracy of original YOLOv5s model is 89.2%on 1500 images containing four types of cracks.After introducing the attention mechanisms,the accuracy of the model improved by 5.7%and 7.1%respectively,reaching 94.9%and 96.3%.The results indicate that the improved YOLOv5s algorithm can achieve automatic detection and classification of bridge cracks and has broad application prospects in practical bridge performance testing.