DisasterViT: A Lightweight Vision Transformer for Disaster Classification

Le résumé fourni par la source

Natural disasters can cause catastrophic damage, and often strike unexpectedly, leaving communities vulnerable and disrupting critical infrastructure. Response authorities must quickly classify the disaster to mitigate its effects, which can be challenging because of transportation and communication breakdowns in affected areas. Unmanned Aerial Vehicles (UAVs) can be used for real-time surveillance in remote regions, capturing critical visuals that enable Computer Vision (CV) techniques to analyze and classify the disaster type, supporting timely response efforts. For the classification task, we proposed DisasterViT a lightweight Vision Transformer (ViT) with an External Attention (EA) mechanism, optimized for high performance. DisasterViT was trained on a custom dataset of 3,720 images from four distinct disaster categories: cyclone, earthquake, flood, and wildfire, achieved an impressive overall accuracy of 91% in classifying diverse disaster scenarios. The code is publicly available at: https://github.com/shamastabraiz/disaster_vit.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.