Research on knowledge graphs construction based on geospatial data

Geospatial data present significant challenges for knowledge integration and semantic modeling due to their inherent high dimensionality, heterogeneity, and multi-source nature. Leveraging the advantages of structured semantic networks, knowledge graphs have emerged as effective tools for representing entities and their semantic relations. This paper proposes a comprehensive framework for constructing geospatial knowledge graphs, promoting the transformation of geographic data from “information” to “knowledge” and ultimately to “wisdom”. First, this paper leverages semantic segmentation models based on deep learning to automatically extract geographic features (e.g., buildings, roads or water systems) from remote sensing imagery, compensating for the limitations of traditional data sources like OpenStreetMap (OSM) and Point of Interest (POI) in terms of geometric accuracy and update frequency. Subsequently, it integrates multisource heterogeneous data to build unified entity models that encapsulate geometric, attribute, and relation information. Building upon these advancements, this paper takes Xi’an, Shaanxi, China as a case study to construct a real-world geospatial knowledge graph, thereby demonstrating its effectiveness in intelligent retrieval, spatial reasoning, and related tasks.