Road extraction through Yangwang-1 nighttime light data: A case study in Wenzhou, China

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Artificial Intelligence

Background:

• Road extraction from remote sensing is crucial for urban planning and transportation management.
• Daytime satellite imagery has been the traditional source, but Nighttime Light (NTL) data offers new potential.
• NTL data for road extraction is an emerging research area.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for road extraction using NTL imagery.
• To address the gap in research concerning NTL data for road network mapping.
• To enhance the capabilities of road extraction from satellite data.

Main Methods:

• A refined U-Net model incorporating Cross-Attention Mechanisms (CA U-Net) was developed.
• The model was trained and tested using Yangwang-1 NTL images from Wenzhou City.
• Performance was compared against Support Vector Machines (SVM) and Optimal Threshold (OT) methods.

Main Results:

• The CA U-Net model achieved a high F1 score of 84.46% for road extraction.
• The proposed model significantly outperformed SVM and OT methods in accuracy.
• The study demonstrated the effectiveness of the developed model in complex urban environments.

Conclusions:

• NTL data, specifically from Yangwang-1, is a viable and reliable source for road extraction.
• Deep learning frameworks, like the proposed CA U-Net, are effective for road extraction tasks using NTL data.
• This research advances the utility of NTL data for comprehensive road network mapping and analysis.