LIGHTWEIGHT DEEP LEARNING WITH MOBILENETV2 FOR CLASSIFYING VIRAL DISEASES IN TOMATO PLANTS

Abstract

Tomato plants are highly susceptible to viral diseases, which significantly
reduce yield and threaten global food security. Traditional diagnostic techniques
are labor-intensive, time-consuming, and prone to error, highlighting the need
for automated and scalable solutions. This research develops and evaluates a
lightweight deep learning model using the MobileNetV2 architecture to classify
tomato leaf images into four categories: Tomato Yellow Leaf Curl Virus
(TYLCV), Tomato Spotted Wilt Virus (TSWV), Tomato Mosaic Virus
(ToMV), and healthy leaves. A dataset of 10,000 labeled images obtained from
the PlantVillage repository was preprocessed using normalization, resizing, and
data augmentation. The model was trained and tested using TensorFlow and
Keras, achieving an overall accuracy of 97.8%, with precision, recall, and F1-
scores exceeding 95% across all classes. The results demonstrate that
MobileNetV2 provides an efficient and accurate solution suitable for mobile and
edge devices, enabling early disease detection and improved decision-making
for farmers. These findings underscore the potential of lightweight CNN
architectures in precision agriculture and contribute toward sustainable,
technology-driven crop management.