Comparison of Artificial Intelligence Techniques for the Detection of Pest Diseases in Agriculture

Abstract

Tomato is one of the important crops that are produced in large quantities and have a high commercial value. In recent years, deep learning has had unprecedented results in many applications, specifically convolutional neural networks (CNNs), in plant disease classification. In this paper, we conduct a comparison of different Convolutional Neural Network (CNN) algorithms—AlexNet, VGG16, and Inception V3—to find the optimal algorithm for classifying tomato leaves as healthy or diseased. Both the VGG16 and Inception V3 architectures achieved an equal accuracy of 99.55% for the training data, while the AlexNet architecture had an accuracy of 97.76%. However, there is a difference in the accuracy of the test data; VGG16 excels with an accuracy of 90%, Inception V3 has an accuracy of 89%, and AlexNet has an accuracy of 75%. These tests were conducted on the PlantVillage Database, which consists of 32,545 leaf images (one healthy and ten disease classes). The proposed system can be utilized in tomato fields for the early detection of diseases to avoid production loss.