Bridge Crack Detection Based on Attention Mechanism

Abstract

With the strong support of the country for bridge construction and the increase in supervision of the safety of old bridges, the visual-based bridge crack target detection has a problem of incomplete target framing due to the characteristics of the bridge crack target, reflecting the current algorithm model's poor ability to accurately identify targets. In this paper, YOLO V5 algorithm was used to address the issue of poor accuracy in bridge crack target detection, and a relevant bridge crack detection dataset was created. Three attention mechanisms, SENet, ECALayer, and CBAM, were respectively fused to improve the model's feature fusion part, and comparative experiments were conducted. The experimental results show that the improved algorithm has increased from 80.5% to 87% in mAP50-95 indicators compared to the original algorithm.