A Hybrid Model for Disaster Damage Detection Using Satellite Images

Efficient and accurate assessment of disaster-induced damage is paramount for effective disaster response and relief operations. Traditional methodologies reliant on manual analysis are often impractical for large-scale disaster areas. The challenges inherent in disaster damage assessment, such as varying environmental conditions, noise in satellite imagery, and the need for rapid response, are confronted through a multistep approach. We propose a novel approach leveraging the combined capabilities of EfficientNet and HRNet architectures for disaster damage detection in dual-temporal satellite imagery. By harnessing the unique strengths of these models, we aim to improve building localization and multi-level change detection. Extensive experimentation on a comprehensive building damage assessment dataset (xBD) showcases that our approach achieves state-of-the-art results. Based on our experiments, it was found that this method demonstrated the capability to detect disaster damage with an accuracy rate of 98.2%. This method could be applied to many real-world applications, including Emergency Response Planning and Management, Environmental Monitoring and Conservation, and Agricultural Damage Assessment, so that it could be applied in many different contexts.