Research on the Spatial Distribution of Wudang Mountain Historical and Cultural Heritage Architecture Based on Convolutional Neural Networks and Geographic Information Systems
Description
This dataset provides the complete spatial and attribute data supporting a study on the spatial distribution of the Wudang Mountain Ancient Architectural Complex, a UNESCO World Cultural Heritage site in China. It comprises 15 scanned thematic maps covering building periods, preservation status, building types, land use, elevation, road networks, and visibility, along with extracted attribute tables for identified heritage buildings. The data also include cross‑map matching results generated by three independent algorithms (K‑Nearest Neighbors, DBSCAN density clustering, and spatial indexing), with the spatial indexing method yielding the most comprehensive set of matched point pairs linking the same building entities across different thematic maps. The dataset further contains spatial analysis outputs derived from a hybrid framework that combines Convolutional Neural Networks with attention mechanisms and Geographic Information System tools. These outputs include spatial association rules revealing directional clustering tendencies and dominant building types, K‑means clustering results dividing the matched points into spatially distinct groups, and ArcGIS‑based buffer zones around core roads (ancient pilgrimage paths and scenic roads), kernel density surfaces identifying primary concentration hotspots, and elevation profiles. The analyses demonstrate significant spatial heterogeneity, with certain period groups showing planned dispersion while others exhibit localized clustering that may indicate regional conservation risks, and validate a 200‑meter road buffer as a critical zone for protection priority. This dataset is suitable for comparative spatial studies of other heritage sites, development and validation of machine learning methods for cultural heritage image analysis, conservation planning and risk zoning, and educational use in digital heritage and GIS courses. Users should note that coordinates derived from scanned maps are relative image coordinates rather than absolute geographic coordinates, thus analytical methods based on relative positional relationships are preferred over models requiring high precision. The data are provided in common formats (PNG and tabular data) compatible with major GIS and statistical software.