Enhancing road damage identification in satellite images through synthetic data
Journal
International Journal of Disaster Risk Reduction
Journal Volume
116
Start Page
105091
ISSN
2212-4209
Date Issued
2025-01
Author(s)
Lien-An Chen
DOI
10.1016/j.ijdrr.2024.105091
Abstract
Assessing road damage in the aftermath of disasters is crucial for timely and effective responses to emergencies. Traditional assessment methods, which rely on civilian reports and on-site surveys, pose challenges because of their time-consuming and labor-intensive nature, particularly during widespread disasters. This research addresses the limitations of existing road damage identification methods employed in satellite imagery by emphasizing the use of local damage features to enhance the performance of deep learning models. We introduce a novel approach that leverages specific regional characteristics of Haiti and Indonesia to train our models. By incorporating local damage features, our methodology facilitates the differentiation between damaged and undamaged roads and improves the model's adaptability to diverse disaster scenarios. Additionally, a road damage recognition framework is developed, which combines a road extraction model and a Siamese damage identification model through an integrated post-processing strategy. Our results indicate a significant enhancement in the model's capability to classify road damage when trained using locally relevant features, particularly in the test regions. Future research will focus on generating and evaluating synthetic damage features across diverse lighting conditions, color tones, and terrains to further improve the robustness of damage identification models.
SDGs
Publisher
Elsevier BV
Type
journal article