Automatic Clustering and Classification of Coffee Leaf Diseases Based on an Extended Kernel Density Estimation...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Current plant disease classification methods are limited by training data and dataset specifics.
• Manual sample collection and labeling are labor-intensive, prone to errors, and may overlook minor symptoms, leading to misclassification.

Purpose of the Study:

• To develop a fully automated framework for leaf disease diagnosis.
• To improve classification accuracy by clustering symptoms non-parametrically.
• To reduce the reliance on large-scale datasets for classifier training.

Main Methods:

• A modified color process extracts regions of interest.
• Syndromes are self-clustered using extended Gaussian kernel density estimation and nearest neighborhood probability.
• Each symptom cluster is independently presented to a ResNet50 classifier.

Main Results:

• The proposed extended Gaussian kernel effectively clusters neighboring lesions into single symptom groups.
• Equal priority is given to all clusters by the ResNet50 classifier, minimizing misclassification.
• The framework achieved up to 98% accuracy on coffee leaf disease datasets.

Conclusions:

• The automated framework offers an efficient and accurate solution for plant disease diagnosis.
• Symptom clustering significantly reduces classification errors and the need for extensive training data.
• This approach enhances the reliability of automated plant disease identification systems.