Crack Detection and Displacement Measurement of Earth-Fill Dams Based on Computer Vision and Deep Learning

Weiwu Feng, Siwen Cao, Lijing Fang, Wenxue Du, Shuaisen Ma

Intelligent crack detection and displacement measurement are critical for evaluating the health status of dams. Earth-fill dams, composed of fragmented independent material particles, are particularly vulnerable to climate changes that can exacerbate cracking and displacement. Existing crack segmentation methods often suffer from discontinuous crack segmentation and misidentification due to complex background noise. Furthermore, current skeleton line-based width measurement techniques demonstrate limited accuracy in processing complex crack patterns. To address these limitations, this study introduces a novel three-step approach for crack detection in earth-fill dams. Firstly, an enhanced YOLOv8-CGA crack segmentation method is proposed, incorporating a Cascaded Group Attention (CGA) mechanism into YOLOv8 to improve feature diversity and computational efficiency. Secondly, image processing techniques are applied to extract sub-pixel crack edges and skeletons from the segmented regions. Finally, an adaptive skeleton fitting algorithm is developed to achieve high-precision crack width estimation. This approach effectively integrates the pattern recognition capabilities of deep learning with the detailed delineation strengths of traditional image processing. Additionally, dam crest displacements and crack zone strain field are measured via the digital image correlation (DIC) method. The efficacy and robustness of the proposed method are validated through laboratory experiments on an earth-fill dam model, demonstrating its potential for practical structural health monitoring (SHM) applications in a changing climate.