Application of image processing and deep learning in crack detection of historical structures: a systematic review

Crack detection is essential for the structural safety assessment and conservation of historical structures. Recently, image processing technique (IPT)-based and deep-learning (DL)-based methods have been increasingly applied to automated crack inspection due to their non-contact operation, high efficiency, and intelligent analytical capabilities. This paper presents a systematic review of image-based crack detection technologies for historic structures based on the PRISMA framework. The review summarizes recent developments in crack identification, segmentation, and geometric quantification using both IPT- and DL-based approaches. The findings indicate that IPT-based methods remain advantageous for interpretable geometric analysis and multi-source sensing integration, whereas DL-based methods demonstrate superior automation and feature-learning capabilities under complex background. However, current studies are still constrained by limited heritage-specific datasets and domain shift. This review further discusses the relationship between crack segmentation quality and the reliability of geometric parameter estimation and highlights future research directions for intelligent crack assessment in heritage conservation.