A Method for Extracting High-Resolution Building Height Information in Rural Areas Using GF-7 Data.

Area of Science:

• Geospatial Science
• Remote Sensing
• Artificial Intelligence

Background:

• Accurate building height data is crucial for disaster management, damage assessment, population modeling, and urbanization studies.
• Existing methods for extracting building height are limited, especially in rural areas of China.
• The Gaofen-7 satellite (GF-7) provides high-resolution imagery suitable for detailed mapping.

Purpose of the Study:

• To develop and validate a novel method for extracting building height in rural areas using GF-7 satellite imagery.
• To assess the accuracy of the proposed method against reference LiDAR data.
• To promote the application of satellite data for large-scale building height surveys in underserved regions.

Main Methods:

• A combined approach of photogrammetry and deep learning was employed.
• A deep learning model, DELaMa (based on LaMa architecture), was developed for digital surface model (DSM) editing.
• Building height was determined using the percentile value of the normalized digital surface model (nDSM) within building footprints.

Main Results:

• The DELaMa model effectively edited DSMs, preserving topographic details and predicting topography within building masks.
• Extracted building heights showed high consistency with ICESat-2 LiDAR data.
• Validation metrics included R² of 0.83, Mean Absolute Error (MAE) of 1.81 m, and Root Mean Square Error (RMSE) of 2.13 m.

Conclusions:

• The proposed photogrammetry and deep learning method accurately extracts building heights from GF-7 data in rural areas.
• This approach significantly enhances the utility of satellite imagery for large-scale building height mapping, particularly in rural environments.
• The findings support improved disaster management and urban planning through more comprehensive geospatial data.