Skip to content
vvangfayePublic

About

[RSE 25] Official implementation of the paper mKGR.

Resources

Stars

23 stars

Watchers

2 watching

Forks

Latest commit

 

History

29 Commits

Folders and files

Repository files navigation

mKGR

The official code of Remote Sensing of Environment paper Learning to Reason over Multi-Granularity Knowledge Graph for Zero-shot Urban Land-Use Mapping.

image

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

Pipeline

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.

image

Other Code

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.

Contact

If you have any questions about it, please let me know. (Create an 🐛 issue or 📧 email: wangfaye@whu.edu.cn)

About

[RSE 25] Official implementation of the paper mKGR.

Resources

Stars

23 stars

Watchers

2 watching

Forks

Releases

Packages

Used by

Contributors

Languages