Detecting Dust on Solar Panels using Deep Learning Models

الملخص

           Dust accumulation on solar panels is a well-known factor that reduces energy production and, if left unaddressed, can significantly affect system efficiency. Early and accurate detection of dust is therefore essential for effective maintenance and sustained power generation. This study presents an automated approach for identifying dust on solar panel using deep learning–based image classification. Two convolutional neural network (CNN) architectures, Xception and VGG16, were evaluated to assess their effectiveness in this task. To improve model robustness and generalization, an extensive preprocessing pipeline was applied to the dataset. The experimental results demonstrate that the Xception model achieved superior performance, reaching a classification accuracy of 98.82%, while the VGG16 model attained an accuracy of 93.49%. Beyond performance evaluation, this work emphasizes model interpretability. The integrated gradients method was employed to generate visual explanations of the models’ predictions, highlighting image regions that contributed most to the classification outcomes. These visual interpretations provide valuable insight into the decision-making process and enhance confidence in the results. Overall, the findings indicate that deep learning offers a reliable solution for automated solar panel monitoring, reducing reliance on manual inspection. Moreover, incorporating interpretability techniques improves transparency and supports practical deployment. This study contributes to renewable energy maintenance research and lays the groundwork for intelligent systems that integrate automated detection, cleaning strategies, and predictive maintenance to enhance the long-term efficiency of solar power installations