Enhancing Solar Energy Efficiency: A Deep Learning Approach for Dust Detection

In recent decades, power generation from PV systems has become increasingly popular. However, several environmental variables, such as dust deposition on PV panels, have significantly reduced PV energy production. Several approaches are under investigation to identify dust and decrease power loss. These include thermal imaging, image processing, sensors, and IoT devices, machine learning, and deep learning. The orientation of a solar panel and the angular inclination relative to the horizontal plane affect one of the environmental factors that impact its performance: since they have an impact on how much solar energy that enters into the photovoltaic panel's surface. Dust, for one, is another example of environmental factors: the quantity of energy output can be influenced by coming into contact with dust. Dust particles lying on the panel's surface have the potential to reduce light from reaching the solar modules, which would drastically reduces the energy output. Cleaning or reducing modules is critical, but these procedures must be continuously monitored and assessed for their effects. Image processing algorithms can be used to detect dust to optimize solar panels. In this paper, a novel dataset from Kaggle comprising images of dusty and clean solar panels is presented, aiming to facilitate the research on automated dust detection algorithms. The dataset is utilized to evaluate the performance of existing state-of-the-art (SOTA) image classification methods. Subsequently, a new Convolutional Neural Network (CNN) architecture, ResNet-50 with an additional attention mechanism, is proposed, specifically designed for solar panel dust detection. The experimental results shows that the proposed method outperforms other SOTA algorithms, achieving an accuracy level of 96.2%.