|
Multi-level damage detection of buildings after natural disasters using deep learning models and satellite image analysis
|
Milad Farmahini Farahani   , Houshang Dolatshahi Pirooz * |
| School of Civil Engineering, University of Tehran, Tehran, Iran. |
|
|
Abstract: (14 Views) |
Objective: Natural disasters such as earthquakes, floods, and hurricanes cause extensive damage to buildings and urban infrastructure. Rapid and accurate post-disaster damage assessment plays a crucial role in crisis management, optimal allocation of relief resources, and reconstruction planning.
Method: In this study, using the XBD dataset and the deep learning-based models YOLOv8 and YOLOv11, a multi-level classification framework was developed to detect and categorize building damage. The models were evaluated using Accuracy, Precision, Recall, F-score, mean Intersection over Union (mIoU), and Dice Coefficient metrics.
Results: YOLOv11 outperformed YOLOv8, achieving an Accuracy of 0.82, F-score of 0.81, mIoU of 0.73, and Dice of 0.86. Both models successfully identified and distinguished three damage levels — minor, severe, and complete — across various disaster types.
Conclusion: Further analysis indicated that the type of disaster and the degree of urban density significantly influenced model performance, with reduced accuracy in densely built areas, particularly for minor damages. The superior performance of YOLOv11 is attributed to its optimized architecture and the Dynamic Anchor Head mechanism, which enhances its capability to capture complex spatial damage patterns. Overall, the proposed deep learning framework provides an effective, automated approach to post-disaster building damage assessment, supporting the development of intelligent crisis management systems and improving the efficiency of resource allocation during disaster response.
|
|
| Keywords: Natural disasters, damage detection, remote sensing, feature extraction, deep learning. |
|
|
Full-Text [PDF 1377 kb]
(6 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2025/09/27 | Accepted: 2025/11/2 | Published: 2025/11/2
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
