Federated Learning for Urban Energy Efficiency: Detecting Thermal with UAV-based Imaging and AI

The fast and accurate localization of heat loss areas to increase energy efficiency can be improved by automating the detection of thermographic anomalies through UAV-based
thermal imaging combined with Deep Learning (DL). However, there are still challenges such as data sharing, resource constraints and privacy concerns in urban environments. Federated
Learning (FL) offers a solution by enabling privacy-preserving model training on decentralized devices, making it suitable for resource-constrained applications. Using NVFlare and the
U-NET segmentation model, we investigate Federated Learning (FL) for thermal hot spot detection based on urban features. The FL clients are selected based on the geographic location
of the image and help district heating network operators detect leakages in underground pipelines via false alarm removal.