Large vision model-based crack segmentation under limited supervision

Crack segmentation is an important basis for automatic evaluation of the apparent state of bridge structures. The emergence of large vision models has revolutionized the basic process of object segmentation and demonstrated strong generalization ability. In this paper, a novel large vision model-based crack automatic segmentation method under limited supervision is proposed. By adding a crack segmentation head, the segmentation generalization ability of the large vision model is retained, and the automatic segmentation of cracks is realized. The effectiveness and accuracy of the proposed method are illustrated by comparing with the state-of-the-art crack segmentation methods on eight commonly used crack datasets.