Risk Evaluation of Crack Disease in Helankou Rock Painting Based on Deep Learning

This study investigates the deterioration risks affecting the Helankou rock paintings, with a focus on the segment extending from Shanhu to Shuiguan in Helankou Town. A quantitative analysis of fissure distribution characteristics — covering ten key parameters such as orientation, number of fissure sets, and spacing — was conducted using the statistical window method. To address the challenge of rapid and accurate risk evaluation, a hybrid machine learning model was developed by integrating Random Forest (RF), Multilayer Perceptron (MLP), and the Sine-Cosine Algorithm (SCA). The resulting SCA-optimized MLP-RF model achieved a prediction accuracy of 92%, outperforming the standalone MLP (76%) and RF (84%) models. The model serves as the core of a rapid risk assessment framework informed by the vulnerability evaluation of disaster-bearing entities. A risk classification matrix was further employed to integrate risk and vulnerability outcomes, enabling efficient spatial zoning of risk levels across the study area. The vulnerability assessment revealed that 47.4% of the rock paintings are already damaged, with deterioration concentrated in the core heritage zones. Risk zoning results indicate that high-risk areas account for 13.1%, and medium-risk areas for 10.8%, primarily located on the rock painting surfaces and within a 5 cm radius. These findings provide valuable insights into the spatial progression of fissure-related deterioration and offer a scientific basis for localized risk management and targeted conservation strategies. The proposed assessment framework enhances the precision and efficiency of heritage risk evaluation and contributes to the long-term preservation of rock art under environmental and structural stressors.