Deep Learning-Based Multi-Class Tomato Leaf Disease Classification for Precision Agriculture

Abstract

Diseases affecting tomato leaves substantially decrease crop output and quality, resulting in economic losses if not identified quickly. Manual inspection is labour-intensive and sometimes imprecise, making it inappropriate for extensive agricultural operations. This research presents a deep learning framework using DenseNet-121 for the automatic multi-class classification of tomato leaf diseases. The model was trained and validated using the PlantVillage dataset, which contains more than 54,000 labelled pictures of tomato leaves classified into Early Blight, Late Blight, Leaf Mold, and Healthy categories. Training utilized RMSprop optimization, categorical cross-entropy loss, and data augmentation methods such as rotation, flipping, and brightness modifications to enhance generalization. The proposed approach achieved an accuracy of 99.17%, precision of 98.34%, and F1-score of 98.33%, surpassing baseline models. In addition to numerical performance, the system facilitates practical applications: it may be included in mobile applications, IoT-based monitoring systems, or cloud platforms, enabling farmers to identify illnesses in real time, implement early treatments, and mitigate crop losses. This study illustrates the capabilities of AI-enhanced precision agriculture, providing a scalable and dependable approach for sustainable crop management.