Utilizing the YOLO Framework for Unnecessary Particle Identification on Solar Panel Surfaces

Solar cells as a renewable energy source are essential for alleviating fossil fuel dependence and transforming to a more sustainable energy future. Unnecessary particles that fall on the surface of the solar panel block the sunrays, thereby reducing the efficiency and overall performance of the panel. Over time, the accumulation of such objects may cause the reduction of output power, especially in highly polluted regions. This paper aims to automatically detect various classes of unwanted particles, such as dirt, dry leaves, and dry tree branches, on pre-installed solar cells using You Only Look Once (YOLO)-based predictive models. The precision value variation increases with confidence, reaching 1.00 at a confidence value of 0.94. On the other hand, the recall value of the model remains moderate, up to a threshold value of 0.8, which controls the ability to find false instances of classes. Training loss curves decrease with increasing epochs, while validation loss decreases quickly, indicating the model's improvements in generalising to unseen data.