An image segmentation-based quantitative method for assessing corrosion risk levels in bronze artifacts

Bronze artifacts hold significant historical and artistic value but are highly susceptible to corrosion. Although scientifically predicting corrosion risk is essential for conservation, current manual evaluations are subjective, inefficient, and lack standardized quantitative criteria. This limitation hinders digital heritage efforts. To address this issue, we propose an image-based framework that uses a fine-tuned SAM2 model to automate the segmentation of corrosion regions. Guided by deterioration patterns, the framework extracts four key indicators: area ratio, color factor, aggregation index, and shape complexity. A hybrid subjective-objective weighting scheme integrates these indicators to predict risk levels. Experimental results demonstrate robust segmentation performance (mIoU = 0.81, mDice = 0.90). When applied to bronze artifacts excavated in Guangzhou, the framework produced risk evaluations that closely match expert assessments. These findings confirm the feasibility and robustness of the framework for automated corrosion monitoring and highlight its significant practical potential.