Tomato Leaf Disease Detection: A Comparative Study using Deep Convolutional Networks and Vision Transformer

The detection of plant diseases is crucial for mitigating economic losses and enhancing agricultural productivity. Convolutional Neural Networks (CNNs) have proven highly effective in automating plant disease identification from images of leaves, fruits, and stems. Traditional CNN architectures, such as AlexNet, Inception, ResNet, DenseNet, and EfficientNet, have demonstrated high accuracy in plant disease classification tasks. Recent advancements, including Vision Transformers and lightweight architectures for IoT applications, have further expanded classification capabilities. In this study, we e valuate the performance of three deep CNN architectures, EfficientNet B0, InceptionV3, and ViT b16 in classifying plant diseases. We employ transfer learning to leverage pre-trained models, optimizing computational efficiency while p reserving l earned f eatures. Our evaluation focuses on identifying the trade-off between model size and classification accuracy, aiming to determine the most suitable architecture for mobile and resource-constrained environments. Results indicate that EfficientNet B0 offers a more compact and efficient solution while maintaining high accuracy, making it a promising choice for real-world agricultural applications.