Automated Ground Truth Generation for Learning-Based Crack Detection on Concrete Surfaces

This article introduces an automated data-labeling approach for generating crack ground truths (GTs) within concrete images. The main algorithm includes first-round GTs, pre-training a deep learning-based model, and second-round GTs. On the basis of generated GTs training data, detection model can be trained in self-supervised manner. pre-trained is effective after it re-trained using contribution this study proposal GT generation process at pixel level. Experimental results show that are similar to manually marked labels. Accordingly, cost implementing methods reduced significantly because data labeling by humans not necessitated.