DVCS-VIT: Disaster Victim Classification System Using Vision Transformer

Finding survivors among wrecked homes and debris, especially after an earthquake, requires early identification and a quick, precise reaction from the Search and Rescue (SAR) team. Disaster victim classification (DVC) is a critical component of rapid evacuation for victims, traditionally dominated by convolutional neural networks (CNNs). However, the advent of transformer-based architectures and large-scale pretrained models has significantly shifted the trend by offering enhanced performance and efficiency. So, in this work, ViT was adopted to design a system for classifying victims of disasters by the pre-trained model (ViT-B16). Then, this system integrated with a communication protocol upon intelligent machines to detect victims in disaster cases. The model achieved a 99.78% test accuracy and a 0.0125 test loss, indicating strong generalization capabilities on unseen data. This demonstrates superiority over existing related works, most of which traditionally relied on CNNs that achieved accuracies of 81.82% [14],44.90% [24], and 78% [25].