A novel hierarchical framework for plant leaf disease detection using residual vision transformer

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate and timely detection of plant leaf diseases is vital for crop yield and food security.
• Plant diseases stem from various factors including pathogens and environmental conditions.
• Existing detection methods may lack efficiency or accuracy.

Purpose of the Study:

• To develop an advanced model for early and accurate detection of plant leaf diseases.
• To enhance the extraction of critical disease-related features while optimizing computational resources.
• To improve the overall efficiency and effectiveness of plant disease identification systems.

Main Methods:

• A novel hierarchical residual vision transformer model was proposed.
• The model integrates an improved Vision Transformer with the ResNet9 architecture.
• It was trained and evaluated on diverse datasets: Local Crop, Plant Village, and Extended Plant Village.

Main Results:

• The proposed model demonstrated superior performance in identifying plant leaf diseases across 13, 38, and 51 classes.
• It achieved higher accuracy compared to established models like InceptionV3, MobileNetV2, and ResNet50.
• The model efficiently extracts discriminating details with reduced parameters and computations.

Conclusions:

• The developed hierarchical residual vision transformer offers a robust solution for early plant leaf disease detection.
• This advancement contributes to safeguarding agricultural productivity and global food security.
• The model's efficiency and accuracy present a significant improvement over existing methods.