This study introduces a lightweight detection model for drone-based solar panel contamination monitoring, leveraging the MobileViT architecture with a hybrid CNN-Transformer design and localized self-attention mechanisms. The model reduces parameters by 6%, computational overhead by 8.5%, and memory usage by 6.1%, while achieving a validation accuracy of 84.1%. Preprocessing techniques like Gaussian filtering and histogram equalization enhance image quality and feature extraction. The model outperforms DenseNet, ViT, and Swin Transformer, offering a 7% improvement in test accuracy over DenseNet with 54% lower FLOPs. Designed for real-time operation, it provides a scalable, cost-effective solution for photovoltaic maintenance, maximizing energy output.