Exploring Generative AI for YOLO-Based Object Detection to Enhance Flood Disaster Response in Malaysia

$F$ lood disasters pose significant threats to communities around the world. In Malaysia, floods are recurrent and severe natural disasters due to its geographical location. Focusing on the response phase of flood disaster management, this study identifies the critical issues involved, including the limitations of manual observation in traditional methods and the lack of structured flood image databases. This research explores the use of generative artificial intelligence (gen-AI) to create synthetic images as an alternative source in flood disaster detection and response. By adopting a systematic approach with the You Only Look Once (YOLO) algorithm, the model is trained on gen-AI-produced synthetic images and tested using real drone-captured images from past flood events. The model demonstrates promising preliminary results with the highest performing model, achieving a mean average precision (mAP50) of 96.7%, mAP@[0.5,0.95] of 78.7% and an F1 score of 90.8%. These findings highlight the potential of gen-AI in supporting flood disaster response efforts in Malaysia, aiming to integrate the mix of synthetic images during non-flooding periods and actual drone images during flood scenarios into the comprehensive FloodIntel framework for future research.