Urban Waterlogging Disaster Assessment Based on Vision Transformer

With the accelerated pace of urbanization, the frequency and intensity of extreme rainfall events in cities are increasing, leading to frequent flooding disasters. Accurately estimating water accumulation information during floods is crucial for effective disaster prevention and mitigation. This study proposes an urban waterlogging disaster prediction model based on Vision Transformer (ViT), categorizing waterlogging severity into four levels: level_O (no waterlogging), level_l (mild waterlogging), level_ 2 (moderate waterlogging), and level_ 3 (severe waterlogging). The model was trained and tested using 2,230 images collected from the internet, along with data augmentation operations. Experimental results demonstrate that the model achieves an average recognition accuracy of 92.79% across all categories, with outstanding performance in level_ 0 and level_3 classifications. This high-accuracy model provides an effective technical means for rapid assessment and response to urban waterlogging disasters, significantly enhancing urban flood disaster management efficiency and emergency response capabilities.