During a disaster, quick and accurate assessment of the situation is required. However, information at disaster sites is fragmented, and collecting and organizing relevant data is time-consuming. Therefore, we propose a system that automatically summarizes, classifies, and visualizes damage situations based on disaster images posted by citizens. In this system, citizens upload disaster images with location information using a web form and the captions of the damage situation are generated from the images. The system uses a large-scale language model to automatically classify images under different damage categories from the generated captions and visualize them on a disaster map. This enables real-time structuring of images posted by citizens and intuitive understanding of disaster sites.