Abstract
Geospatial information systems can face challenges from incomplete data, such as missing locations for building entrances, and difficulties processing complex natural language queries. A system can dynamically identify geospatial features using a multi-stage process. For example, a large language model can interpret a user's natural language query and an associated image to identify a relevant point of interest. Subsequently, a specialized machine learning segmentation model can analyze a localized region of the image to delineate a specific feature, such as an entrance. This approach can facilitate the on-demand generation of geospatial feature coordinates in response to semantic requests, which can be used to supplement existing data stores and potentially improve data completeness for downstream applications.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Leung, Mira, "System for Identifying Geospatial Features Using Language Model-Guided Segmentation", Technical Disclosure Commons, (July 07, 2026)
https://www.tdcommons.org/dpubs_series/10821