A Deep Learning-Based Framework for Tomato Leaf Disease Classification Using Image Preprocessing and Convolutional Neural Networks

Abstract

Tomato leaf diseases can reduce crop productivity and quality, motivating automated image-based classification systems. This paper presents an experimentally evaluated convolutional neural network (CNN) framework for ten-class tomato leaf disease classification. The final reported model is a custom CNN consisting of three convolutional layers, three max-pooling layers, ReLU activations, and a fully connected classification head with ten outputs. Convex-hull preprocessing and Otsu thresholding are applied before CNN classification. The dataset contains 18,345 images covering nine disease classes and one healthy class, with 300 images reported as the held-out test set. The experimental record reports a class-wise average test accuracy of 94.5%. Class-wise precision, recall, and F1-score are additionally reported from the supplied experimental figures. The submitted training/validation learning curve spans approximately 1000 epochs, which resolves an inconsistency in the earlier manuscript that stated five epochs. The comparison with VGG16, InceptionV3, and MobileNet is retained as a descriptive author-run baseline comparison; because the available record does not preserve complete split and hyperparameter information for every baseline, the values are not treated as a rigorously controlled benchmark. The study demonstrates performance on the evaluated public dataset, but it does not establish generalization to independently collected field images