Precision diagnosis of tomato diseases for sustainable agriculture through deep learning approach with hybrid data augmentation

Abstract

Tomato is a key crop in global agriculture, yet it faces yield and quality challenges due to various diseases. Traditional disease identification methods are slow and require expertise, limiting their practicality in large-scale farming. Integrating automated disease detection with precision agriculture provides a timely, accurate diagnosis, promoting sustainable practices. However, the scarcity of real-world data hampers effectiveness. To address this issue, data augmentation techniques simulate variations in farm images, enriching datasets for improved detection of diseases. This investigation aims to identify seven different tomato diseases, such as bacterial spot, early blight, late blight, and others, while also detecting healthy plant leaves. Unlike previous studies that relied on the controlled PlantVillage dataset, this study utilizes the real-world PlantDoc dataset. The study addresses different challenges faced throughout the model development process, like data scarcity and imbalances. A hybrid data augmentation technique is introduced to increase the dataset size from 737 images to 6696 images, which improves the accuracy and robustness of the computer vision model. The study employs the YOLOv8n deep convolutional neural network, achieving 96.5 % mAP, 97 % precision, 93.8 % recall, and 95 % F1 score. The results demonstrate a significant improvement in disease detection, addressing challenges from inadequate datasets and advancing AI-driven precision agriculture. The proposed YOLOv8n model has the potential to be applied beyond its current scope by training it on datasets of other crops. The model can learn and generalize the unique image features associated with various crop types, expanding its utility in agricultural applications. This flexibility allows the model to detect and classify plant characteristics, diseases, or pests across different crops, enabling its use in diverse agricultural environments. As a result, the YOLOv8n model could serve as a robust tool for precision farming, helping to optimize crop management and enhance productivity on a broader scale.