Monitoring Subsidence Area with the Use of Satellite Radar Images and Deep Transfer Learning

Area of Science:

• Geosciences
• Remote Sensing
• Artificial Intelligence

Background:

• Land subsidence poses significant risks to populated areas, impacting human life and property.
• In situ measurements for monitoring large-scale subsidence are often impractical due to cost, time, and accessibility issues.
• Differential Interferometry Synthetic Aperture Radar (DInSAR) offers a powerful solution for monitoring land deformations with high accuracy and resolution.

Purpose of the Study:

• To develop and evaluate an automated system for detecting and monitoring subsidence troughs.
• To apply deep-transfer learning techniques for enhanced subsidence analysis.
• To compare the performance of deep learning models against traditional methods like Hough and circlet transforms.

Main Methods:

• Utilized deep-transfer learning, specifically convolutional neural networks, for automated detection of subsidence troughs.
• Applied the method to the Upper Silesian Coal Basin, an area with significant mining-induced subsidence.
• Compared the automated detection results with those obtained using Hough transform and circlet transform algorithms.

Main Results:

• Deep-transfer learning demonstrated effectiveness in identifying subsidence troughs in interferograms.
• The automated detection system shows promise for improving the efficiency and accuracy of subsidence monitoring.
• Performance comparison provided insights into the strengths of deep learning over traditional image processing techniques for this application.

Conclusions:

• Automated detection of subsidence troughs using deep-transfer learning is a viable and efficient approach.
• This technology can significantly aid in the development of early warning systems for land subsidence.
• Further research can refine these methods for broader application in geohazard monitoring.