CUGUV: A Benchmark Dataset for Promoting Large-Scale Urban Village Mapping with Deep Learning Models

Area of Science:

• Geographic Information Systems (GIS)
• Remote Sensing
• Urban Planning

Background:

• Accurate mapping of urban villages (UVs) is vital for urban planning and policy support.
• Satellite imagery offers an efficient alternative to field surveys for UV mapping.
• Existing research lacks comprehensive UV maps for diverse cities, hindering model transferability.

Purpose of the Study:

• To address the scarcity of diverse urban village data for model development and validation.
• To create a benchmark dataset for evaluating and enhancing the robustness of UV mapping models.
• To develop an innovative framework for effective cross-city urban village mapping.

Main Methods:

• Curated a benchmark dataset (CUGUV) with thousands of urban village samples from 15 Chinese cities.
• Developed an integrated framework leveraging multiple data sources for cross-city UV mapping.
• Evaluated model performance using accuracy, precision, and F1-scores.

Main Results:

• The CUGUV dataset provides a diverse collection of urban village samples for research.
• The proposed framework achieved over 92% in overall accuracy, precision, and F1-scores.
• The developed model significantly outperforms existing state-of-the-art methods in large-scale UV mapping.

Conclusions:

• The CUGUV dataset and the proposed framework enhance the understanding and modeling of urban villages.
• This work improves the reliability and transferability of urban village mapping models.
• The findings contribute to better large-scale urban village mapping and urban planning.