SOLAR PANELS CONTAMINATION DETECTION USING CNN

Solar panels, as an important source of renewable energy, are exposed to external factors such as dust accumulation and environmental pollution, which reduces their efficiency. To ensure timely cleaning of panels from contamination, monitoring of their condition is required. Monitoring can be carried out directly by a person or using visual detectors. This paper examines the second case, namely the use of AI to assess panel contamination. The AI used to assess the condition of the panels may be trained on images captured by cameras with a different resolution than the surveillance system in which it is embedded, which may result in the AI's inability to produce results under new conditions. This work evaluates the impact of changing the quality of images processed by AI on its effectiveness. To do this, a CNN (Convolutional Neural Network) model was used and a set of images of clean and dirty solar panels was taken, followed by their separation by resolution. The model was trained on a set of images of one quality and tested on others, and this approach was applied for each resolution. As results were obtained, elements from other sets were added to each set to stabilize the training of the neural network. During the work, the importance of using images of different qualities for training a neural network was assessed, as well as the minimum set to prevent overtraining due to the lack of quality differences. Taking into account the results obtained, it is possible to prevent the creation of inflexible neural networks, as well as save on using the minimum required data to stabilize training.