Image Classification of Wheat Rust Based on Ensemble Learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Wheat rust diseases, including stem and leaf rust, pose significant threats to global crop yield.
• Manual identification of wheat rust is labor-intensive and prone to errors, necessitating automated solutions.

Purpose of the Study:

• To develop an automated and accurate method for identifying wheat rust diseases using ensemble learning.
• To improve upon existing convolutional neural network (CNN) models for wheat rust detection.

Main Methods:

• An ensemble learning (WR-EL) method was developed, integrating multiple CNN models (VGG, ResNet 101, ResNet 152, DenseNet 169, DenseNet 201).
• Techniques such as bagging, snapshot ensembling, and stochastic gradient descent with warm restarts (SGDR) were employed.
• A novel SGDR-S algorithm was proposed to further optimize performance.

Main Results:

• The WR-EL method demonstrated substantial accuracy improvements over individual CNN models, with increases ranging from 8% to 32%.
• The proposed SGDR-S algorithm enhanced F1 scores for healthy, stem rust, and leaf rust wheat by 2%, 3%, and 2%, respectively, compared to the standard SGDR algorithm.
• The study achieved high accuracy in distinguishing between healthy wheat, stem rust, and leaf rust.

Conclusions:

• The WR-EL method offers a highly accurate and efficient approach for wheat rust identification.
• This automated system can facilitate timely disease management, mitigating economic losses and improving wheat production.
• The findings contribute to advancing precision agriculture through AI-driven crop disease diagnostics.