A general and simple automated impervious surface mapping approach based on three-dimensional texture features (3DTF)...

Area of Science:

• Environmental Remote Sensing
• Geospatial Analysis
• Sustainable Development

Background:

• Mapping impervious surfaces is crucial for environmental management and sustainable development.
• Existing remote sensing methods often require extensive training samples, limiting their applicability.

Purpose of the Study:

• To develop and evaluate an automated, training-sample-free method for mapping impervious surfaces.
• To assess the method's performance across diverse geographical and environmental conditions.

Main Methods:

• Utilized texture features (Contrast, Gabor wavelets, Con_Gabor) from Sentinel-2 imagery.
• Constructed three-dimensional texture features (3DTF) for impervious surfaces.
• Employed a K-means classifier for automated impervious surface mapping.

Main Results:

• Achieved high overall accuracies (e.g., >91% in China, >89% in arid regions).
• Demonstrated comparable performance to Artificial Neural Network (ANN) and Random Forest (RF) methods.
• Validated the robustness of the 3DTF feature combination across bands, seasons, and sensors.

Conclusions:

• The automated 3DTF-based method is effective and efficient for mapping impervious surfaces.
• This approach offers a valuable, sample-free alternative for environmental monitoring and planning.
• Further research is needed for complex areas using high-resolution imagery.