To solve the problems of low accuracy, high false detection rate, and high missed detection rate in current target detection methods for apparent diseases in concrete bridges, an improved YOLOv5s method is proposed. To achieve more effective fusion of features at different scales and increase receptive fields, an improved spatial pyramid pooling module is added to the YOLOv5s network to enhance feature extraction capabilities and reduce computational cost; a light-weight attention module is incorporated into the YOLOv5s network to tackle the high false detection and missed detection rates caused by the cross-distribution of different defect features in disease images; and a loss function considering vector angles is adopted to solve the problems related to varying defect sizes, classification difficulties and small dataset-induced boundary box regression mismatches. Experimental results show that the improved YOLOv5s detector significantly improves accuracy while reducing false detection and missed detection rates in the task of detecting apparent diseases in bridges.