Advances in crack dataset development and deep learning-based detection models

Yunlong Song, Qi Zhang, Yumeng Su, Shiying Zhang, Ruilin Wang, Weiping Zhang, Zhuo Bi, Youling Yu,
Advances in crack dataset development and deep learning-based detection models,
Journal of Building Engineering,
Volume 116,
2025,
114734,
ISSN 2352-7102,
https://doi.org/10.1016/j.jobe.2025.114734.
(https://www.sciencedirect.com/science/article/pii/S2352710225029717)
Abstract: This paper is a review of recent advances in crack detection using deep learning. Crack detection is vital for structural safety and service life, especially in enabling intelligent maintenance and resilience of buildings. The paper summarizes commonly used crack datasets and compares those used for classification, object detection, and segmentation tasks. It further analyzes annotation methods, evaluation metrics, loss functions, and representative deep learning architectures. Performance evaluation, post-processing, 3D modeling, and visualization strategies are also discussed, along with practical applications in roads, bridges, and buildings. The review identifies key challenges at the data, model, and system levels, and proposes engineering-oriented evaluation metrics. Finally, it highlights future research needs in addressing data scarcity, crack variability, and limited model generalization, and advocates for weakly supervised learning, multi-modal fusion, and interpretable AI to enhance system reliability and practicality. This review identifies limitations of existing datasets, emerging trends in deep learning models, and future directions for practical deployment in building inspection.
Keywords: Crack detection; Deep learning; Image segmentation; Object detection; Structural health monitoring