MSLU-100K: A Large Multi-Source Dataset for Land Use Analysis in Major Chinese Cities

Area of Science:

• Geographic Information Science
• Remote Sensing
• Urban Planning

Background:

• High-quality land use datasets are crucial for research but challenging to create due to complexity and spatial heterogeneity.
• Existing datasets may lack sufficient detail or quality for advanced land use classification and recognition tasks.

Purpose of the Study:

• To introduce MSLU-100K, a novel, large-scale, multi-source land use dataset for China.
• To provide a high-quality resource that addresses the limitations of current land use datasets.

Main Methods:

• Developed a human-computer collaboration framework for dataset construction.
• Integrated multi-source data, including remote sensing and Point of Interest (POI) data.
• Employed a multi-level classification approach combining manual labeling and deep learning for quality assurance.

Main Results:

• MSLU-100K contains over 100,000 irregular parcel samples from 81 Chinese cities.
• Parcels are categorized into 7 primary and 28 secondary land use types.
• Over 57% of the dataset achieved high-quality classification (Levels 4 and 5), significantly boosting performance.

Conclusions:

• MSLU-100K is a robust and high-quality dataset for land use recognition.
• The dataset supports advancements in urban planning and spatial research.
• The human-computer collaboration and multi-level classification approach ensure data reliability.