目标检测, 裂缝检测, YOLO, 数据增强, 无人机

Most surface cracks of concrete dams are still detected manually, which is risky and inefficient. With the development of information technology, intelligent detection of surface cracks of concrete dams is the further development direction. Problems such as few dam crack samples, large ratio of crack length and width as well as complex background environment are difficult to solve for the existing algorithms using image recognition. Aiming at the above-mentioned problems, based on the open source database and YOLOv5 model, technical means including transfer learning, data augmentation and segmentation annotation are adopted. As a result, automatic detection of surface cracks of concrete dams is realized, with mAP@0.5 reaching 0.942, and F1- score reaching 0.919. Compared with traditional algorithms, the algorithm proposed in this paper has better generalization, accuracy and real-time performance. The experiment shows that the algorithm can also be applied to the concrete crack detection for bridges, roads and other projects after proper adjustment. Moreover, the surface image can be collected by UAV, camera, vehicle mounted camera or other equipment, which allows more image acquisition methods.

object detection, crack detection, YOLO, data augmentation, UAV