Digital preservation methods at an advanced level should protect rock paintings because they need to resist damage while remaining accessible to the public. The lack of accurate documentation using traditional sketching and photography techniques, alongside information loss, prevents adequate conservation progress. This research establishes an all-inclusive rock painting data acquisition technology and database construction framework, which improves rock art data precision, accessibility, and durability. The new system utilizes high-resolution imaging together with LiDAR scanning and multispectral analysis for creating detailed representations of rock art pictograms. The preprocessing phase enhances data quality through a noise reduction routine of 28%, along with contrast enhancement to achieve element segmentation for uniform image clarity. The pattern recognition process utilizes Vision Transformers (ViTs) and Swin Transformers, which produced recognition results at 87.3% accuracy and 91.5% accuracy, respectively. The database system uses spatial indexing along with AI retrieval functions to deliver fast search performance at 120.5 ms for ViT and 86.4 ms for Swin Transformer. High-quality image processing is verified by measured Peak Signal-to-Noise Ratio (PSNR) at 32.4 dB and Structural Similarity Index (SSI) at 0.92. Study results from various rock painting sites confirmed that the proposed method provides superior accuracy and search speed, together with better conservation outcomes than traditional methods. AI-based feature extraction combined with transformer architectures and structured data storage proves beneficial for rock painting preservation and analysis in the research findings. This digital framework offers an effective solution for heritage conservation with classification and restoration capabilities, which promotes academic research, along with cultural heritage management development.
Jun Nie (Thu,) studied this question.