Multi-kernel inception-enhanced vision transformer for plant leaf disease recognition.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate plant disease identification is vital for crop yield and food security.
• Manual identification is labor-intensive and requires specialized expertise.
• Existing computer vision models often struggle with real-world field conditions.

Purpose of the Study:

• To develop a robust deep learning architecture for plant disease identification.
• To address the limitations of current models in handling diverse image conditions (lab vs. real-field).
• To improve the accuracy and efficiency of automated plant disease diagnosis.

Main Methods:

• Proposed an Inception-Enhanced Vision Transformer (IEViT) architecture.
• IEViT integrates local and global feature extraction capabilities.
• Utilized multiple filters with varying kernel sizes for efficient feature learning.
• Conducted experiments on five diverse datasets, including lab and real-field images.

Main Results:

• The IEViT model demonstrated superior performance compared to state-of-the-art deep learning models.
• Achieved high accuracy rates across multiple datasets: 99.23% (apple leaf), 99.70% (rice), 97.02% (i-bean), 76.51% (cassava leaf), and 99.41% (plantvillage).
• The proposed architecture achieved these results with fewer parameters than existing models.

Conclusions:

• The Inception-Enhanced Vision Transformer (IEViT) is a robust and efficient architecture for plant disease identification.
• IEViT effectively handles variations in image conditions, performing well on both laboratory and real-field data.
• This approach offers a promising solution for automated, accurate, and scalable plant disease diagnosis in agriculture.