Relative-Breakpoint-Based Crack Annotation Method for Lightweight Crack Identification Using Deep Learning Methods

After years of service, bridges could lose their expected functions. Considering the significant number and adverse inspecting environment, urgent requirement for timely efficient inspection solutions, such as computer vision techniques, have been attractive in recent years, especially those bridge components with poor accessibility. In this paper, a lightweight procedure apparent-defect detection is proposed, including crack annotation method detection. First all, order to save computational costs improve generalization performance, we propose herein relative-breakpoint build instance segmentation dataset, critical process supervised vision-based method. Then, trained models based on classic Mask RCNN Yolact are transferred evaluate effectiveness proposed To verify correctness, universality generality crack-detection framework, approximately 800 images used model training, while nearly 100 saved validation. Results show that can achieve level 90% both accuracy recall values, limited dataset.