Framework for Cross-Disaster Building Damage Assessment Using Cost-Sensitive Learning

Abstract

Rapid and reliable assessment of structural damage following disasters is critical for prioritizing rescue operations. In this study, we present a unified deep learning framework for building damage assessment from satellite imagery. The proposed approach integrates segmentation-driven feature construction, morphological processing, cost-sensitive learning, and Cross-Disaster strategies to enable robust performance under limited and imbalanced data conditions. Our approach combines an adapted U-Net for building segmentation with a hybrid CNN-DNN classifier for damage evaluation, and incorporates cross-learning to assess generalization across different disasters and imaging conditions.We evaluate our method on the xBD and BRIGHT datasets, leveraging both pre-disaster optical and post-disaster SAR imagery. Despite extremely limited and imbalanced samples, our framework achieves competitive performance, with a macro F1-score of 70\% on the Mexico earthquake (xBD) and up to 98\% on earthquake cases in BRIGHT. Cross-validation and cross-disaster transfer learning further demonstrate the model’s generalizability and resilience, highlighting its potential for aiding real-time disaster response.Although constrained by data availability, our results indicate that lightweight, cost-aware deep learning models can provide actionable insights for resource-constrained rescue operations. This work highlights both the promise and the challenges of deploying AI for real-time disaster response.