<![CDATA[The buildup of dust on solar panels can greatly diminish energy output, lower system efficiency, and raise operational expenses. A productive way to tackle this problem is to utilize image classification through Convolutional Neural Network (CNN) techniques. This study examines the classification capabilities of four CNN models, namely MobileNetV2, DenseNet121, InceptionV3, and Xception, using transfer learning. These models leverage pre-trained weights from large datasets such as ImageNet to accelerate convergence and improve generalization. The dataset of images utilized in this research is obtained from Kaggle and includes pictures of both clean and dusty solar panels. The dataset was divided into training, validation, and testing subsets using a stratified approach to ensure balanced class distribution across all subsets. During training, class weighting was used to address potential class imbalance. The models were developed using TensorFlow with multi-GPU support, optimized using the AdamW optimizer, and fine-tuned to enhance performance. Model evaluation was conducted using accuracy, precision, recall, and F1-score metrics. Among all the architectures evaluated, the Xception model achieved the best performance with an accuracy of 90.52%, outperforming MobileNetV2 with an accuracy of 87.92%, DenseNet121 with 89.78%, and InceptionV3 which achieved 87.73%. These results indicate that modern CNN-based models can effectively recognize relevant visual patterns to detect dust on solar panels.]]>
Copyrights © 2025
