Deep-learning-based sub-meter urban construction-site mapping reveals China's dual-track urban renewal

Area of Science:

• Earth and Environmental Sciences
• Urban Planning and Development
• Remote Sensing and Geospatial Analysis

Background:

• China's New-Type Urbanization Plan, the world's largest, requires precise monitoring.
• Urban construction sites (UCSs) are key indicators of urban expansion and renewal.
• Existing monitoring methods lack the high-precision, fine-scale detail needed.

Purpose of the Study:

• To develop a deep-learning framework for nationwide, high-resolution mapping of urban construction sites.
• To enable detailed progress monitoring of China's urbanization.
• To analyze urbanization patterns and their environmental impacts.

Main Methods:

• A sub-meter-resolution deep-learning framework was developed.
• Weakly supervised pixel-level UCS annotation using the Segment Anything Model.
• A spectral-texture dual-branch segmentation network achieved 94.3% overall accuracy.

Main Results:

• Identified 541,177 UCSs across 372 cities, including micro-sites.
• K-means clustering revealed a dual-track urbanization pattern (incremental expansion + stock optimization).
• UCS construction correlates with PM₂.₅ concentrations, with limited effectiveness of dust-proof nets.

Conclusions:

• The proposed framework offers a 'microscope' for evaluating new-type urbanization.
• Findings support sustainable urban planning and environmental management.
• Identified urbanization patterns differ from classical models, highlighting a dual-track approach.