Rapid post-disaster infrastructure damage characterisation using remote sensing and deep learning technologies: A tiered approach

This study presents a tiered, multi-scale approach to infrastructure damage characterization post-disaster, specifically focusing on bridges. Given the strategic importance of bridges and the challenges posed by limited access during hostilities or disasters, the study integrates remote sensing and deep learning technologies to automate and accelerate the damage assessment process. The approach utilizes Sentinel-1 Synthetic Aperture Radar (SAR) images, crowdsourced data, and high-resolution images for deep learning-based automatic damage detection. The methodology applies interferometric coherence differences and semantic segmentation to enhance the accuracy of damage detection at regional and component scales. This approach was validated through a case study of 17 bridges in Ukraine that were damaged due to human-induced interventions. The study highlights how technology can streamline decision-making, support rapid restoration, and contribute to enhancing the resilience of critical infrastructure.

Takeaways:

1. Tiered Approach for Scalable Damage Assessment: A multi-scale tiered methodology is proposed to efficiently assess damage from the regional to the component level, enabling quick identification of critical infrastructure damage.
2. Integration of Remote Sensing and Deep Learning: The use of Sentinel-1 SAR images and high-resolution images in combination with deep learning enables automated, accurate damage detection and characterization, reducing reliance on manual assessments.
3. Enhanced Damage Characterization with Semantic Segmentation: The integration of semantic segmentation techniques improves the reliability of damage detection, making it possible to identify structural changes in bridges more accurately.
4. Real-time and Crowdsourced Data: The use of crowdsourced data alongside satellite imagery enhances the timeliness and coverage of damage assessments, facilitating faster decision-making in post-disaster scenarios.
5. Application to Conflict Zones: The approach is particularly valuable in conflict zones, like Ukraine, where access to affected areas may be limited, offering an automated solution to monitor and assess infrastructure damage remotely.
6. Support for Restoration and Resilience: The methodology aids in accelerating restoration efforts and improving infrastructure resilience, which is crucial for recovery after human-induced or natural disasters.