Disaster image tagging using generative ai for digital archives

Disaster digital archives play a crucial role in preserving and disseminating data on various natural disasters. To manage these archives effectively, images need to be tagged appropriately and comprehensively, and machine learning is expected to handle this task efficiently. However, existing machine-learning tagging models fail to extract detailed disaster-related information. This study focuses on extracting disaster-specific tags from images using the latest machine-learning techniques. More specifically, we use generative AI to create descriptions of images and extract tags from these descriptions, allowing for more detailed information retrieval compared to traditional tags. By including prior information that the images are disaster-related in the prompts, we aim to achieve more specialized disaster tagging. Qualitative evaluation results suggest that the proposed method extracts more disaster-related tags than existing tagging models, indicating that it provides effective tags for users searching disaster images. This study applies real-world test cases using images from the 2011 Tohoku Earthquake and the 2016 Kumamoto Earthquake.