Related Experiment Video
Updated: Jul 1, 2026
08:27
Image Recognition and Parameter Analysis of Concrete Vibration State Based on Support Vector Machine
Published on: January 5, 2024
Deep learning-based classification of earthquake-damaged buildings using terrestrial images.
Hamed Kashani1, Amirmohammad Sahebzadeh2, Mahdi Naimi Jamal2
1Center for Infrastructure Sustainability and Resilience Research, Department of Civil Engineering, Sharif University of Technology, Iran.
Disasters
|June 30, 2026
Summary
This study developed a deep learning model using ground-level images to quickly classify earthquake building damage. The model achieved 93.5% accuracy, aiding rapid disaster response and urban resilience.
Area of Science:
- Computer Science
- Civil Engineering
- Disaster Management
Background:
- Effective post-earthquake building damage assessment is crucial for rescue and resource allocation.
- Ground-level terrestrial images offer detailed insights often missed by aerial or satellite imagery.
Purpose of the Study:
- To develop and evaluate a deep learning classifier for rapid, reliable assessment of earthquake-induced building damage using terrestrial images.
- To categorize buildings into 'not damaged,' 'damaged,' or 'collapsed' for immediate post-event triage.
Main Methods:
- A ResNet50-based deep learning model was trained on a curated dataset of post-earthquake building images.
- The model utilized terrestrial imagery to capture ground-level damage details.
Main Results:
- The deep learning classifier achieved an overall accuracy of 93.5% in validation.
- Demonstrated the feasibility of fast, building-scale damage assessment.
Conclusions:
- This terrestrial image-based deep learning approach complements existing remote sensing methods for post-earthquake analysis.
- Accelerates decision support for disaster response, improving urban resilience and potentially saving lives.