Deep Transfer Learning for Tomato Leaf Diseases Detection and Classification using Pre-trained Models

Tomatoes are among the most important and widespread crops worldwide. The amount of tomatoes produced varies according to the method of fertilization used. Leaf diseases are the key reason affecting both the quantity and quality of crop yields. Consequently, it is important to diagnose and classify these diseases in the early stages. Recent advances in convolutional neural networks (CNNs) have demonstrated excellent performance in solving image diagnosis and classification tasks in computer vision. Transfer Learning offers an efficient and cost-effective approach to building deep CNNs. In this work transfer learning approaches were applied to accurately classify the tomato leaf diseases. VGG16, VGG19, and ResNet50 models were tuned up and customized for transfer learning. Plant Village Dataset was used in this research work which consists of a total of 16,012 images belonging to 9 disease classes and 1 healthy class. The results indicate that VGG 16, when used in conjunction with transfer learning, outperformed both VGG19 and Resnet50, achieving a superior training accuracy of 95.1%.