Tomato Leaf Disease Classification Using a Mobile-Integrated Deep Learning Framework Based on Custom CNN and ResNet50 Models

Olayiwola Charles Adesoba; Waliu Olalekan Apena; Folasade Mojisola Dahunsi; Ayodele Martins Ajayi

Olayiwola Charles Adesoba Federal University of Technology, Akure
Waliu Olalekan Apena Federal University of Technology, Akure
Folasade Mojisola Dahunsi Federal University of Technology, Akure
Ayodele Martins Ajayi Federal University of Technology, Akure

Keywords: Tomato diseases, Deep learning, TDD-Net, Mobile diagnosis, Precision agriculture

Abstract

Tomato leaf diseases pose a significant threat to crop productivity and require timely diagnosis to minimize yield losses and improve disease management. This study developed and validated a mobile-integrated deep learning framework for automated tomato leaf disease classification using a custom convolutional neural network (TDD-Net) and a fine-tuned ResNet50 model. A heterogeneous dataset comprising 15,549 publicly available and field-acquired tomato leaf images representing bacterial spot, early blight, grey leaf spot, late blight, and healthy leaves was used for model development. Both models were trained under identical optimization settings and evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). The proposed TDD-Net achieved superior performance, attaining a validation accuracy of 98.94%, precision of 0.9882, recall of 0.9863, F1-score of 0.9871, and an AUC of 0.9944, outperforming the fine-tuned ResNet50 across all evaluation metrics. The trained TDD-Net model was subsequently integrated into a FlutterFlow-based mobile application with Firebase support to enable real-time disease diagnosis using smartphone images. The results show that a task-specific convolutional neural network provides highly accurate tomato leaf disease classification while supporting practical deployment through an accessible mobile decision-support system for precision agriculture.