AI-Enhanced Detection of Thermal Anomalies in Urban Roofs via Drone-Assisted Infrared Thermography (UAV-IRT)

Abstract This study investigates automated detection of rooftop thermal anomalies using Unmanned Aerial Vehicle-Based Infrared Thermography (UAV-IRT) and deep learning object detection models. UAV thermal images are used to identify potential thermal bridges on building rooftops. The experiment is framed as a preliminary workflow validation performed on a subset of the Karlsruhe reference dataset. The study evaluates four YOLO models, including YOLOv9, YOLOv10, YOLOv11, and YOLOv12. The results indicate that all models can detect major thermal anomaly regions in UAV thermal imagery. YOLOv9 achieves slightly higher detection accuracy in the tested dataset. These results demonstrate the feasibility of the proposed workflow under reference-dataset conditions, while validation on a dedicated historic-building dataset remains necessary before claiming applicability to heritage or historic urban environments.