Precision Agriculture: Automated Classification of Potato Leaf Diseases using Transfer Learning on Vision Transformers

Potato crops are a cornerstone of global food security, yet they remain highly vulnerable to diseases like Early Blight and Late Blight, which can devastate yields if not identified early. This technical report explores the development of an automated diagnostic tool using a specialized niche classifier developed within a Jupyter Notebook environment. By leveraging the Potato Leaf Disease dataset from Kaggle, we implemented a transfer learning strategy using a pre-trained Vision Transformer (ViT) architecture. The methodology involved freezing the foundational layers of the model—originally trained on general imagery—and fine-tuning a custom classification head to recognize the specific cellular patterns and "bullseye" concentric rings characteristic of potato pathogens. To ensure the model could handle real-world variability in field photography, we applied data augmentation techniques including random rotations and color jittering. The resulting model achieved a high validation accuracy, demonstrating a sophisticated ability to distinguish between visually similar fungal infections that often confuse human observers. This project underscores how high-performance machine learning models can be adapted for specific agricultural purposes with minimal computational overhead, providing a scalable blueprint for mobile-based crop monitoring and precision farming.