Bridge Crack Detection Method Based on Yolov5-Dense Attention Multi-Scale Fusion

作者：Shuai Tan, Jiayi Yang · 发表于：2024 7th International Conference on Mechatronics and Computer Technology Engineering (MCTE) · 年份：2024 · DOI：10.1109/mcte62870.2024.11117851

Bridge crack images have irregular distributions and small crack widths. To improve the accuracy of bridge crack detection algorithms, this paper proposes a bridge crack detection method based on Yolov5-Dense Attention Multi-Scale Fusion (Yolov5-DAMF). The proposed model uses Dense Units to connect the output of each layer with the output of all previous layers, reusing features. This approach optimizes the frequent issues of missed detections and feature information loss in the original model. Using Convolutional Block Attention Module (CBAM) in the Conv-Batch Normalization-SiLU (CBS) module adapts to learn channel and spatial attention weights, highlighting crack locations and suppressing background noise. This improves the accuracy of crack detection. Furthermore, merging the upsampled high-resolution feature map with the shallow feature map allows the model to retain and utilize key information in the input image, improving the detection effect of bridge cracks with small crack widths. Experimental results show that the mAP@0.5 of Yolov5-DAMF is 86.2%, and the mAP@0.5:0.95 is 53.7%, achieving precise bridge crack detection.