Optimizing YOLOv5 model configuration for enhanced accuracy in solar panel dust detection

Abstract Dust accumulation on photovoltaic (PV) panels significantly degrades energy conversion efficiency, posing a critical challenge to solar energy production. While deep learning (DL) methods surpass traditional techniques for dust detection, there is a persistent need for solutions that offer high accuracy in real-time, non-invasive applications. This study addresses this gap by proposing and evaluating a dust detection model based on the YOLOv5 architecture. A custom dataset of solar panel images, categorized by varying levels of soiling, was developed to train and test the model for practical deployment. Experimental results validate the model’s effectiveness, demonstrating high accuracy and robustness in identifying dust accumulation. The findings indicate that the YOLOv5-based approach is a practical and efficient tool for automated solar panel monitoring, enabling timely maintenance to optimize performance and address the limitations of existing detection systems.