Enhanced Maize Leaf Disease Detection and Classification Using an Integrated CNN-ViT Model

Maize crop productivity is significantly impacted by various foliar diseases, emphasizing the need for early, accurate, and automated disease detection methods to enable timely intervention and ensure optimal crop management. Traditional classification techniques often fall short in capturing the complex visual patterns inherent in disease-affected leaf imagery, resulting in limited diagnostic performance. To overcome these limitations, this study introduces a robust hybrid deep learning framework that synergistically combines convolutional neural networks (CNNs) and vision transformers (ViTs) for enhanced maize leaf disease classification. In the proposed architecture, the CNN module effectively extracts fine-grained local features, while the ViT module captures long-range contextual dependencies through self-attention mechanisms. The complementary features obtained from both branches are concatenated and passed through fully connected layers for final classification. Data from Mendeley and Kaggle were used to build and check the model, and the model did this by applying image resizing, data normalization, expanding its training data, and shuffling the data to increase generalization. Additional testing is done on the corn disease and severity (CD&S) dataset, which is separate from the main combined dataset. After validation, the accuracy of the proposed model was 99.15%, and each of its precision, recall, and F1-score equaled 99.13%. To confirm it is statistically reliable, 5-fold cross-validation was performed, reporting on the Kaggle + Mendeley set an average accuracy of 99.06% and on the CD&S dataset 95.93%. As both of these scores are high, it shows that the model works well across other datasets as well. Experiments have shown that Hybrid CNN-ViT works better than standalone CNNs. Dropout regularization and using the RAdam optimizer greatly improved both stability and performance. The model stood out as a reliable, high-accuracy method for discovering maize diseases correctly, which may be valuable in real agricultural settings.