Automated irregularity characterization and restoration prioritization for stone masonry heritage structures using image processing and deep learning

• Achieved high-precision segmentation of stone masonry blocks using the Segment Anything Model, a deep learning architecture, enabling robust morphological analysis of heritage structures; • Quantified morphological irregularities in stone masonry through a novel geometric framework integrating four indices including irregularity of wall shape, distribution uniformity, aggregation dispersion ratio, and a composite wall global index to assess structural heterogeneity; • Validated methodology at the Zhibo Watchtower, a nationally protected cultural relic in China, demonstrating applicability to heritage conservation; • Developed a data-driven priority assessment criterion for restoration, linking the composite wall global index to predictive restoration demand metrics. Stone masonry structures are widely employed in culturally significant heritage sites, including iconic monuments such as the Great Wall and the Zhibo Watchtower. These historical constructions exhibit geometric irregularities stemming from material heterogeneity and non-standardized historical construction practices. To systematically quantify these irregularities, this study builds an integrated novel framework combining image processing with four geometric indices: irregularity of wall shape, distribution uniformity, aggregation dispersion ratio, and a composite wall global index ( WGI ) which is the weighted average of the first three parameters. Leveraging WGI , we developed a priority assessment criterion for guiding restoration efforts in degraded stone masonry walls. High-precision segmentation of individual stone blocks was achieved using the Segment Anything Model, a deep learning architecture, which was integrated with image processing algorithms to extract size parameters and classify stones into four typological categories: standing stones, block stones, strip stones, and flake stones. Validated through a case study at the Zhibo Watchtower—a nationally protected cultural heritage site—the analysis revealed that the masonry composition is dominated by block stones (34.91%) and strip stones (40.07%), with standing stones (12.22%) and flake stones (12.80%) present in smaller proportions. Despite spatial heterogeneity in surface stone distribution, the overall configuration demonstrates a coordinated balance between aggregation and dispersion characteristics. Critically, the WGI exhibited a linear correlation with restoration demand severity, enabling the construction of a priority classification framework to objectively assess structural conservation needs. This methodology provides a systematic, data-driven approach for heritage preservation, bridging gaps between quantitative geometric analysis and practical restoration strategies.