The official code of Remote Sensing of Environment paper Learning to Reason over Multi-Granularity Knowledge Graph for Zero-shot Urban Land-Use Mapping.
Abstract: This paper introduces a multi-granularity knowledge graph reasoning (mKGR) framework. Only with indirect supervision from other tasks, mKGR can automatically integrate multimodal geospatial data as varying granularity entities and rich spatial-semantic interaction relationships. Subsequently, mKGR incorporates a novel fault-tolerant knowledge graph embedding method to establish relationships between geographic units and land-use categories, thereby reasoning land-use mapping outcomes. Extensive experiments demonstrate that mKGR not only outperforms existing zero-shot approaches but also exceeds those with direct supervision. Furthermore, this paper reveals the superiority of mKGR in large-scale holistic reasoning, an essential aspect of land-use mapping. Benefiting from mKGR's zero-shot classification and large-scale holistic reasoning capabilities, a comprehensive urban land-use map of China is generated with low-cost.
- Products: Publicly accessible on ArcGIS Online, download the products on Zenodo.
- Code: Publicly available in this repository.
- Dataset: Publicly available on Zenodo.
- Data-processing pipeline: Fully open-sourced. The complete preprocessing (raw geospatial data → per-city shapefiles + seed labels) is in KG_pre, with the nationwide reproduction data on Zenodo.
Ubuntu 20.04 (or other Linux distribution), one GPU (video memory greater than 12GB and support cuda)
- python>=3.11.5
- numpy>=1.26.2
- pytorch>=2.2.1
- pandas>=2.2.2
- geopandas>=0.14.0
Three stages; the output of each is the input of the next. See each folder's README to run it.
| Stage | Folder | Role |
|---|---|---|
| 1 | KG_pre | raw geospatial data → per-city shapefiles + seed labels |
| 2 | KG_construction | shapefiles → MGKG triplets, id maps, train/valid/test/predict splits |
| 3 | KG_embedding | train the fault-tolerant embedding, infer land-use, export result shapefiles |
Ready artifacts on Zenodo — pick where to start:
KnowledgeGraph.zip(5 cities) — the constructed MGKG; start at stage 3.OriShapefile.zip(5 cities) — per-city shapefiles; start at stage 2.NationalKGpreReproduction.zip(nationwide, 366 cities) — geometry layers to reproduce stage 1.ChinaLandUse.gpkg/China15min.gpkg— the final products.
figure_script: The code for generating the figures in the paper.
landuse_app: The code for 15-minute city application of land-use mapping results.
We have published the land-use mapping and 15-minute walkability results of China on ArcGIS Online.
If you have any questions about it, please let me know. (Create an 🐛 issue or 📧 email: wangfaye@whu.edu.cn)