Advanced neural network architectures for tomato leaf disease diagnosis in precision agriculture

This study addresses the critical challenge of detecting and classifying tomato leaf diseases using advanced deep learning technologies, a pivotal step in enhancing productivity within precision agriculture. We developed a novel diagnostic system capable of categorizing tomato leaves into ten distinct classes—nine corresponding to specific diseases and one for healthy leaves. The system incorporates a dataset of 6,000 images, which underwent extensive preprocessing, including resizing to 256 × 256 pixels, grayscale conversion, normalization, masking, and augmentation, to optimize input quality for the model. Our approach stands out through the integration of state-of-the-art neural network architectures—VGG19, Vision Transformer (ViT), EfficientNetV2, ConvNeXt—and a novel hybrid model specifically designed to leverage the strengths of diverse architectures. Performance evaluation demonstrated that the hybrid model outperformed all individual architectures, achieving an exceptional classification accuracy of 98%, ensuring robust disease detection under varying conditions. To enhance interpretability, Grad-CAM and LIME techniques were employed, highlighting critical image regions and influential features for classification. This comprehensive and innovative approach not only deepens understanding in plant pathology through automated systems but also sets a new benchmark for future advancements in plant disease detection leveraging machine learning.