This dataset focuses on automated photovoltaic (PV) panel detection and fault detection using thermal imagery captured by UAV and includes annotated thermal images of PV panels. The raw thermal images were captured using the DJI Mavic 3T UAV at a photovoltaic farm in Sindos, Thessaloniki. These images were processed to generate a non-overlapping subset of 351 images (640×512) resolution, each containing fully visible PV panel arrays. The PV panel count in each subset can be seen in the following table: Set Images Annotated PV Panels Training 235 18487 Validation 83 5828 Test 35 2363 Total 353 26678 This dataset is intended for training and evaluating deep learning-based object detection models and supports research for renewable energy systems.