Abstract
Plant diseases pose the greatest threat to food supply integrity, and it is a significant challenge to identify plant diseases in their earliest stages to reduce the potential for associated economic damage. Through the use of computer vision, a cutting-edge artificial intelligence is offered as a solution to this problem so that tomato leaf diseases may be classified. The proposed network is expected to provide excellent results. Transfer learning is also used to make the model efficient and cost effective. Since tomato diseases may significantly influence crop output and quality, early identification and diagnosis of these diseases are essential for successful treatment. Deep learning has shown a great deal of promise in plant disease identification, providing excellent accuracy and efficiency. In this investigation, we compared the performance of three different deep learning models-DenseNet169, ResNet50V2, and a transform model, namely ViT, with regard to diagnosing diseases affecting tomatoes. Both diseased and healthy tomato samples were included in the dataset of photos of tomato diseases used for training and testing the models. The DenseNet121 model had the best results, with a training accuracy of (99.88%) and a testing accuracy of (99.00%). This gave it the greatest overall accuracy. Both the ResNet50V2 and VIT models attained high levels of accuracy, with testing accuracies of (95.60% and 98.00%), respectively. Our results demonstrate deep learning's potential for accurate and efficient tomato disease detection, which could aid in early disease management and ultimately improve crop yield and quality. The experimental findings show that the suggested ensemble models stand out due to the short amount of time required for training and testing as well as their exceptional classification performances. Because of this study, professionals will be able to facilitate the early diagnosis of plant diseases in a straightforward and expedient way, thereby preventing the emergence of new infections.