Severity assessment of wheat stripe rust based on machine learning.

Area of Science:

• Plant pathology
• Computational biology
• Agricultural science

Background:

• Accurate wheat stripe rust severity assessment is crucial for understanding pathogen-host interactions, predicting disease spread, and implementing effective control measures.
• Traditional methods for disease assessment can be time-consuming and subjective, necessitating the development of more objective and efficient approaches.

Purpose of the Study:

• To develop and evaluate machine learning-based methods for rapid and accurate severity assessment of wheat stripe rust using image processing.
• To compare the performance of unsupervised and supervised machine learning algorithms in classifying wheat stripe rust severity.

Main Methods:

• Wheat leaf images were processed to segment diseased areas and calculate lesion percentages.
• Training and testing datasets were created using different modeling ratios (4:1 and 3:2), with and without healthy leaves.
• Unsupervised (K-means, spectral clustering) and supervised (support vector machine, random forest, K-nearest neighbor) learning models were trained and tested.

Main Results:

• Both unsupervised and supervised machine learning models achieved satisfactory performance in severity assessment.
• Optimal models, particularly those based on the random forest algorithm, demonstrated excellent performance with 100% accuracy, precision, recall, and F1 scores across all severity classes.
• The inclusion of healthy wheat leaves in the dataset did not significantly alter the high performance of the models.

Conclusions:

• Machine learning offers a simple, rapid, and user-friendly approach for assessing wheat stripe rust severity.
• This study provides a foundation for automated disease severity assessment using image processing technology.
• The developed methods can serve as a reference for assessing other plant diseases.