Early and accurate detection of tomato leaf diseases is critical for improving crop yield and ensuring sustainable agricultural practices. This paper introduces a Hybrid Neuro-Augmented Learning Framework that leverages transformer-based feature extraction, class-aware data augmentation, and knowledge distillation to achieve high classification accuracy with low computational overhead. A Vision Transformer (ViT) serves as a frozen teacher model to guide a lightweight MobileNetV2 student network. Each tomato disease class receives customized augmentation policies to improve model robustness, particularly for visually ambiguous or underrepresented categories. The student network is trained using a combined loss function that balances hard labels and softened teacher logits. Extensive experiments on the PlantVillage Tomato dataset demonstrate that our framework achieves 97.01% accuracy, with consistently high precision and recall across all 10 classes. Grad-CAM visualizations confirm the interpretability of the student model, highlighting its focus on disease-specific leaf regions. This approach offers a practical, scalable solution for real-world agricultural disease monitoring on edge devices, contributing to the advancement of smart farming technologies.