Detection and classification of soiling in PV panels with YOLO algorithm

In this thesis, an artificial intelligence–based image processing and control system was developed to detect and clean surface contamination on solar panels. Pollution on photovoltaic modules—caused mainly by dust, pollen, and bird droppings—significantly reduces the efficiency of solar energy generation. To address this issue, a deep learning–based object detection approach, the YOLOv8 algorithm, was employed to identify and classify the contamination levels on panel surfaces. The detected visual data were analyzed together with operational parameters, and a fuzzy logic–based decision support system was implemented to optimize the cleaning process. In the proposed system, environmental data such as temperature and wind speed were obtained from sensors and processed using fuzzy logic rules. The resulting control signals were used by a microcontroller (Arduino) to operate a servo motor–driven automatic cleaning mechanism. This approach enabled water savings and ensured that the cleaning process was performed only when necessary. Experimental results demonstrated that the YOLOv8 algorithm provides high accuracy in detecting contamination on solar panels, and that its integration with fuzzy logic enhances the system's decision-making efficiency. The proposed hybrid model contributes to improving photovoltaic system performance and reducing maintenance costs, while also offering a practical framework for future smart energy management applications