Semi-Supervised Building Extraction with Optical Flow Correction Based on Satellite Video Data in a Tsunami-Induced...

Area of Science:

• Remote Sensing
• Computer Vision
• Disaster Management

Background:

• Natural disasters are increasing in frequency and intensity globally.
• Accurate building extraction is vital for disaster response, rescue planning, and damage assessment.
• Existing deep learning methods for building extraction struggle with complex disaster scenes and require extensive labeled data.

Purpose of the Study:

• To develop a novel semantic segmentation model for accurate and efficient building extraction from satellite disaster videos.
• To address the challenge of limited labeled training data in post-disaster scenarios.
• To leverage recent advances in satellite video technology for improved building information extraction.

Main Methods:

• A two-part semantic segmentation model comprising a prediction module and an automatic correction module.
• The prediction module utilizes a base encoder-decoder structure with limited instant training data.
• The automatic correction module refines initial predictions using optical flow analysis to correct erroneous semantic information.

Main Results:

• The proposed model demonstrates superior accuracy in building extraction compared to existing methods.
• The method shows improved computational efficiency, particularly in complex natural disaster scenes.
• Effective utilization of limited training data for accurate building segmentation was achieved.

Conclusions:

• The developed model offers an accurate and efficient solution for building extraction in challenging natural disaster scenarios.
• The approach effectively reduces the dependency on large labeled datasets, making it practical for post-disaster applications.
• This research advances the application of deep learning in disaster management through innovative satellite video analysis.