Explainable Hybrid CNN—Transformer Framework for Tomato Disease Diagnosis and Severity Assessment

Abstract

The prompt and precise identification of plant diseases is a critical challenge in contemporary agriculture, as any delay in identification may lead to substantial reductions in crop yields and financial repercussions. Recent developments in deep learning (DL) have facilitated automated plant leaf image disease detection with promising accuracy. However, conventional convolutional neural networks (CNNs) often struggle to concentrate on discriminative disease-affected regions and generalize under varying imaging conditions. Herein, we demonstrate an advanced DL approach for classifying plant diseases using the EfficientNet-B0 backbone integrated with Transformer Encoder blocks for hybrid local-global feature learning to enhance feature representation. We train and evaluate with publicly available plant disease image datasets. Comprehensive experiments have been carried out, and the effectiveness of the proposed model is evaluated with established DL architectures. The model performance is assessed using metrics including accuracy, precision, recall, and F1 score, and confusion matrix analysis. The proposed approach demonstrates superior classification performance and improved interpretability, making it appropriate for practical agricultural disease monitoring applications.