Solar Panel Dust Detection via Grey Wolf Optimization-Based Transfer Learning: A Comparative Study

Solar energy is one of the most important sources of clean and sustainable electricity. However, dust that accumulates on solar panel surfaces over time can significantly reduce their energy output. This study aims to automatically classify solar panel images as clean or dirty using deep learning methods. A publicly available dataset of 1440 solar panel images collected from different regions of Bangladesh was used. Six pre-trained convolutional neural network (CNN) models were employed for feature extraction: DenseNet201, EfficientNetB0, InceptionResNetV2, MobileNet, ResNet50, and Xception. The hyperparameters of the classification layer were optimized using the Grey Wolf Optimization (GWO) algorithm. The models were first evaluated using an 80:20 train/test split, and subsequently validated using 5-fold cross-validation with results reported as mean ± standard deviation. Under 5-fold cross-validation, all six models achieved mean accuracy above 97%, with Xception and MobileNet reaching the highest mean accuracy at 99.93%. These findings demonstrate that transfer learning models combined with metaheuristic optimization can provide a promising solution for automatic dust detection on solar panels.