A Smartphone-Based Detection System for Tomato Leaf Disease Using EfficientNetV2B2 and Its Explainability with...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Tomato diseases cause significant agricultural output and financial losses.
• Timely disease detection is crucial for effective management and mitigation.
• Early detection improves yield, reduces chemical use, and benefits national economies.

Purpose of the Study:

• To develop a precise and effective automated system for identifying various tomato plant diseases.
• To analyze tomato leaf images for disease diagnosis.
• To create a user-friendly application for accurate tomato leaf disease identification.

Main Methods:

• Utilized EfficientNetV2B2 model, a deep learning (DL) architecture, for image classification.
• Employed transfer learning (TF) with pre-existing weights and a 256-layer dense layer for model training.
• Developed a dataset of high-resolution images of healthy and diseased tomato leaves.

Main Results:

• Achieved 99.02% average weighted training accuracy and 99.22% average weighted validation accuracy using 5-fold cross-validation.
• The split method resulted in 99.93% training accuracy and 100% validation accuracy.
• The deep learning approach demonstrated nearly 100% accuracy on a test dataset for tomato leaf disease identification.

Conclusions:

• The developed automated system accurately diagnoses tomato leaf diseases, enabling informed management decisions.
• The system supports sustainable tomato cultivation practices through rapid disease identification.
• Deployment in smartphones and online apps provides accessible and accurate disease diagnosis for users.