Bridge Crack Detection Based on Attention Mechanism

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