Plant Leaves Disease Detection With VIT

Abstract

This work investigates the application of advanced deep learning techniques for the automated detection and classification of tomato plant diseases. The study begins with a comprehensive review of image processing fundamentals in agriculture, tomato plant pathology, and the core concepts of Artificial Intelli- gence, Machine Learning, and Deep Learning, focusing on architectures like CNNs (ResNet50) and Vision Transformers (ViT,DeiT, Swin Transformer). The core methodology involved utilizing the”New Plant Diseases” dataset , implementing data preprocessing and augmentation , and employing K-Fold cross-validation. Four pretrained models ResNet50 ,DeiT 3, SWIN V2, and a Combined ViT+ResNet50 were evaluated based on accuracy, precision, recall, and F1-score. Results indicated exceptional performance across all models, with DeiT 3 achieving the highest accuracy 99.93 .The findings demonstrate the significant potential of deep learning, particularly transformer-based archi- tectures, to advance precision agriculture by providing accurate and efficient tools for plant disease identification.