Researchers from Poland and Spain have developed a transparent framework for autonomous UAV-based inspection of PV installations using thermal and multispectral imaging with AI-based anomaly detection. They utilized a custom hexacopter equipped with FLIR Boson 640 and MicaSense Altum cameras to conduct autonomous flights over residential and industrial PV installations, building a model dataset with various images and manually labeled anomalous modules. Comparing four CNN architectures, they found EfficientNet-B0 to have the best overall performance in detecting anomalies. The system's modular design allows for future enhancements in multi-UAV operation and multi-sensor fusion for high-throughput PV monitoring, as detailed in their study published in Measurement.
https://www.pv-magazine.com/2026/09/03/ai-powered-drone-system-for-autonomous-pv-fault-detection/