Using Densenet169 for Image-Based Classification of Solar Panel Defects

This study investigates the effectiveness of the DenseNet169 deep learning model for the automatic classification of solar panel defects. We used a comprehensive dataset of over $\mathbf{1 1, 0 0 0}$ validated images encompassing six defect categories: bird droppings, dust, electrical damage, physical damage, snow cover, and pristine panels. The model achieves a high training accuracy, demonstrating its ability to learn defect patterns. Validation and test accuracy scores reveal the model’s potential for real-world applications. However, further exploration is warranted to refine the model and address subtle visual similarities between certain defect types, such as physical and electrical damage. This research paves the way for enhanced solar panel inspection techniques, promoting efficient maintenance and optimal energy production.