Towards Sustainable Agriculture: A Transformer-based Hybrid Model for Advanced Leaf Disease Classification

In modern agriculture, early and accurate detection of plant diseases is essential for maintaining crop health and maximizing yield. Traditional methods of disease identification are often labor-intensive and prone to errors, necessitating the development of automated and efficient solutions. Current research in plant disease detection has primarily focused on convolutional neural networks (CNNs), which, despite their success, encounter challenges such as high computational demands and limited generalizability across diverse datasets and environments. This study addresses these challenges by proposing a hybrid model that integrates the Vision Transformer (ViT) for feature extraction with a Support Vector Machine (SVM) for classification. The ViT excels in capturing intricate spatial patterns in leaf images by processing them as sequences of patches. Coupled with the SVM’s robust classification capabilities, this hybrid approach enhances the model’s accuracy while reducing computational overhead. Our methodology uses the Bean Leaf Lesions Classification dataset, which contains 1,167 images. Our proposed model achieved an outstanding accuracy of $\mathbf{9 9 . 2 4 8 \%}$, with precision and recall rates of $\mathbf{0 . 9 9}$ for each class, significantly outperforming existing models. Extensive data augmentation and regularization techniques ensure that the model generalizes well across different datasets, addressing the common issue of overfitting. We also performed hyperparameter tuning to optimize model performance evaluation using accuracy, precision, recall, and $\mathbf{F 1}$-score metrics. The results indicate that our hybrid model provides a highly accurate tool for leaf disease classification and offers practical benefits for deployment in resource-constrained agricultural settings due to its efficiency.