Deep learning-based crack detection method for civil engineering and its multidimensional damage assessment

Civil engineering crack detection faces challenges due to complex environments and external interferences. This paper proposes an improved YOLO v8s-WOMA network, integrating ODConv, C2f-MA modules, and WIoU loss function to enhance crack identification accuracy. A BP neural network is also trained to assess crack damage. Experiments on the CBP dataset compare this method with existing detection algorithms. Results show that the proposed model achieves the highest mAP (90.5%), F1-score (90.3%), and accuracy (89.6%). Bridge crack detection errors remain within 0.1mm (width) and 20mm (length), ensuring precise damage assessment. The model effectively handles complex backgrounds, accurately detects cracks, and meets practical engineering needs.