The Geospatial Technologies for Smart Decisions Research Unit in Morocco has developed an automatic geo-labeling method for UAV thermal imagery. It uses adaptive thresholding and edge refinement with photogrammetric data to detect and localize solar modules. Manual labeling is notrequired. Auto-labeled images were used to train deep learning models for segmentation. Yolov7 achieved the highest mean Average Precision of 98.33 % with a 15 ms inference time. The dataset includes infrared UAV imagery from both ground- and roof-mounted photovoltaic systems. This method reduces data preparation time and supports real-time, on-site inspections. It fits into automated PV monitoring pipelines and enables accurate module-level detection. The research offers a scalable tool for improving large-scale solar system inspections using AI.