Dust detection on solar panels with deep learning

In today's world, solar energy has gained significance as a clean and sustainable energy source. However, among the factors influencing the efficiency of solar panels, the impact of accumulated dirt on the panel surface holds a crucial place. This thesis focuses on the classification of dirty and clean solar panels using transfer learning methods. Determining the cleanliness status of solar panels is of critical importance for optimizing maintenance strategies. In this study, transfer learning methods are examined, specifically aiming to classify dirty and clean solar panels to enhance efficiency in solar energy systems. Dirt accumulation on the surface of solar panels can result from various factors, and the failure to clean this dirt can reduce energy efficiency. Therefore, an automatic system capable of quickly and accurately classifying the condition of solar panels is crucial for optimizing maintenance processes. In this thesis, transfer learning is evaluated for solving a new task with a limited amount of labeled data, using pre-trained deep learning models. Various pre-trained models (EfficientNetB3, ResNet50, MobileNet, VGG19, Xception, InceptionResNetV2, VGG16, ResNet101, DenseNet201, EfficientNetB7) undergo transfer learning methods to create a new model for determining the cleanliness status of solar panels. Models are developed, trained, and tested in the study. Additionally, the accuracy rates of different models are compared, and the models that exhibit the best performance are identified. An analysis is conducted to determine the best model combinations by combining the predictions of specific models and comparing these combined predictions with the actual labels. Thus, a transfer learning approach is established, aiming to achieve a stronger classification result by leveraging the strengths of different models. In conclusion, an effective transfer learning model has been developed in this thesis to detect the cleanliness status of solar panels. The obtained models aim to contribute to optimizing the maintenance processes of solar energy systems, increasing energy efficiency, and supporting sustainable energy use in the long term.