Dam Crack Detection Studies by UAV Based on YOLO Algorithm
Xiang Li, Li Li, Zhigui Liu, Zhangjun Peng, Simin Liu, Shuai Zhou, Xinyao Chai, Kunhong Jiang
Abstract
In this article, a dam crack detection method based on the YOLOv8 algorithm for image collection during unmanned aerial vehicle (UAV) inspections is proposed, aiming to achieve efficient, fast, and accurate visual safety inspections of the dam. The YOLOv8 deep learning framework is used to annotate the public dataset of dam cracks and train the model for dam crack detection. The trained model is then applied to the recognition of UAV inspection images to achieve automatic detection and positioning of dam cracks. The experimental environment for the algorithm is built by capturing high-definition (4000×3000) images of the surface of a concrete structure dam using a DJI Mavic Pro UAV and CMOS camera sensor equipment, followed by foreground-background segmentation of the images and inputting them into the YOLOv8 crack identification model for automatic detection of dam cracks. The experimental results show that this method has high accuracy and efficiency in dam crack detection, with a crack identification accuracy rate and recall rate of 95.2% and 95.8%, respectively. It can effectively reduce labor and time costs while improving detection efficiency and safety.