A deep learning based approach for automated plant disease classification using vision transformer.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Plant diseases significantly reduce agricultural yields, necessitating effective diagnostic tools.
• Farmers require accessible visual information for timely disease identification and management.

Purpose of the Study:

• To develop a real-time automated plant disease classification system using deep learning.
• To evaluate the performance of Vision Transformer (ViT) and Convolutional Neural Network (CNN) models for this task.

Main Methods:

• Implemented a lightweight Vision Transformer (ViT) model for plant disease classification.
• Compared ViT performance against classical CNN methods and hybrid CNN-ViT approaches.
• Trained and evaluated models on multiple plant disease datasets.

Main Results:

• Vision Transformer models demonstrated increased classification accuracy.
• Attention mechanisms in ViT models led to slower prediction times.
• Hybrid CNN-ViT models effectively compensated for the speed reduction caused by attention blocks.

Conclusions:

• Deep learning, particularly ViT, offers a promising approach for automated plant disease identification.
• Hybrid models combining CNN and ViT architectures provide a viable solution for balancing accuracy and real-time performance in agricultural applications.
• Optimized model architectures are crucial for practical deployment in farming environments.