Building Damage Assessment Using Feature Concentrated Siamese Neural Network

Abstract

Rapid and accurate assessment of building damage is essential after natural disasters such as earthquakes, floods, cyclones, and explosions, where timely information supports emergency response, resource allocation, and recovery planning. Conventional inspection methods rely heavily on manual surveys, which are often time-consuming, labor-intensive, and difficult to perform in hazardous environments. To address these limitations, this study presents a FeatureConcentrated Siamese Neural Network (FC-SNN) for automated building damage assessment using pairs of pre-disaster and post-disaster images. The proposed framework employs two identical neural network branches that extract high-level visual representations while emphasizing the most informative structural features through a feature concentration mechanism. By comparing these learned representations, the model effectively identifies variations associated with cracks, collapsed structures, roof damage, façade deterioration, and other forms of structural impact. The extracted features are subsequently analyzed to classify the severity of damage into predefined categories, enabling reliable decision support during post-disaster assessment. The Siamese architecture improves change detection by learning discriminative feature similarities rather than relying solely on pixel-level differences, making the system more robust to variations in illumination, viewpoint, and environmental conditions. Experimental evaluation demonstrates that the proposed approach achieves superior classification accuracy, precision, recall, and F1-score compared with conventional convolutional neural network models while reducing false detections. The framework provides an efficient, scalable, and intelligent solution for automated building damage assessment and has significant potential for integration into disaster management systems, remote sensing platforms, and urban resilience applications.