DGAP-YOLO: A Crack Detection Method Based on UAV Images and YOLO

In the traditional field of safety engineering, cracks on the surface of walls pose a significant threat to buildings. The traditional method of relying on manual visual inspection is inefficient and poses safety risks for certain scenarios such as dams, overpasses, and high-rise buildings. Therefore, it is necessary to research crack detection based on drone inspections. To address this issue, this paper proposes a crack detection algorithm, DGAP-YOLO, based on improvements to YOLOv8. A new attention module called the DGC module is introduced during feature extraction. It combines global context attention mechanisms with methods that enhance spatial and channel interactions, aiding in precise crack detection and localization, thus improving the accuracy and effectiveness of crack detection. Additionally, the original Neck network is replaced with a lightweight lowAFPN for multi-scale feature fusion, enabling the network to more comprehensively capture crack information in images, thereby enhancing the robustness and generalization of the crack detection model. Experimental results on concrete crack images from different scenarios demonstrate that the improved DGAP-YOLO algorithm increases MAP@50 by 4.9%, meeting the performance requirements for crack detection in safety engineering and construction fields and providing a new approach to crack detection.