Hotspot detection on photovoltaic panel using YOLOv5s and YOLOv8s: a deep learning-based analysis

This research paper explores hotspot detection on photovoltaic (PV) panels using two variants of the YOLO (You Only Look Once) algorithm: YOLOv5s and YOLOv8s. Cracking, shading, or soiling may lead to hotspots on solar panels, causing lower power output on a particular spot. Early identification of these hotspots can help maintenance operator avoid more losses while maintaining efficiency. The study uses infrared images to locate hotspots, and the dataset is annotated and split into three sets: training (70%), validation (20%), and testing (10%) for developing the YOLO models. Both models use the same procedures for hotspot detection and continuing tuning to improve the model performance. Both models performed well in terms of precision, recall, F1 score, and mean average precision (mAP), all of which were above 70%. Particularly, YOLOv5s obtained an accuracy of 89.8%, while YOLOv8s attained 88.7%. These findings show both variants of YOLO are useful for detecting and localizing hotspots in PV panels, providing an accurate method for improving solar panel maintenance and performance.