Transfer learning based deep learning model for classifying tomato plant leaf diseases

Abstract Diseases affecting tomato plant leaves pose a significant challenge to global agricultural productivity and food security globally. Traditional manual inspection methods are time-consuming and subjective, necessitating the development of automated techniques. Deep learning has emerged as a promising approach for automating disease classification tasks. In this study, a transfer learning-based deep learning model is developed for classifying diseases in tomato plant leaves. A diverse dataset of images depicting tomato leaves infected with various diseases is taken as input. Four state-of-the-art deep learning models, AlexNet, LeNet-5, DenseNet-121, and InceptionV3, are trained on both imbalanced and balanced datasets. Class balancing experiments are conducted, wherein Experiment 1 involves oversampling minority classes to equalize class sizes, and Experiment 2 balances the dataset around an average threshold using a combination of oversampling and undersampling strategies. Transfer learning is employed to expedite training and improve model convergence. The proposed Tomato Leaf Disease Classification model, inspired by the standard AlexNet, demonstrated superior performance with an accuracy of 95%, recall of 95%, f1_score of 95%, and precision of 95%, compared to other models, particularly on the balanced datasets from Experiment 2. The developed system holds promise for assisting farmers in timely tomato leaf disease classification, thereby facilitating more effective disease management strategies and contributing to sustainable agriculture and food security.