Identification of diseases in tomato leaves using convolutional neural network and transfer learning method

Abstract

The high market demand for tomatoes required high productivity in the agricultural sector. Plant disease is a threat that obstructs tomato production. Disease control is essential to prevent crop failure. Automatic identification is highly recommended for agriculture applications. Inspired by the recent successes research of deep learning for identification, this study applied a computer vision method for identifying tomato plant diseases. This paper adopted a Convolutional Neural Network (CNN) algorithm with the transfer learning approach to identify tomato plant disease. The CNN models such as VVG, ResNet, and DenseNet have been compared to identify and classify tomato plant diseases. The experiments were carried out using a PlantVillage dataset, with 22930 images of tomato leaves diseases and consists of 10 classes. The best model is achieved by ResNet-50 with accuracy, precision, recall, fl-score, and AUC 96.16%, 97%, 96%, 97%, and 97.92%, respectively. The experimental results proved that CNN models could be a useful tool in identifying tomato plant disease.