Plant Disease Classification Using an Ensemble Learning Model of Convolutional Neural Networks and Vision Transformers

Plant diseases pose a significant threat to global agriculture, leading to substantial crop losses and jeopardizing food security. Existing methods for plant disease classification, such as manual inspection, are labor-intensive, error-prone, and require specialized expertise. To address these challenges, this research proposes an integration of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) through an ensemble learning model to enhance the accuracy and robustness of plant disease detection systems. CNNs are proficient in localized feature extraction, while ViTs excel in capturing global contextual information. By combining these complementary capabilities, the ensemble model aims to overcome the limitations of individual architectures, such as CNNs' limited global context understanding and ViTs' computational demands. First we investigated various CNN models suitable for plant disease classification, then ViT model variants are experimented. Using transfer learning, various CNN and ViT models are trained; we call these base models. Finally, prediction outputs of the base models are combined using an averaging ensemble strategy. Experimental results demonstrate the effectiveness of the proposed ensemble model. While standalone models such as DenseNet121 and ResNet50 achieved accuracies of 99.78% and 99.46%, respectively, and ViT-b16 outperformed at 99.80%, the ensemble model surpassed these with an accuracy of 99.87%. This improvement underscores the ensemble framework's ability to integrate CNNs' localized feature extraction with ViTs' global feature understanding, reducing misclassification risks and enhancing reliability.