Robust coffee plant disease classification using deep learning and advanced feature engineering techniques.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Coffee leaf diseases threaten global coffee production and quality.
• Deep learning (DL) shows promise for plant disease identification via image classification.
• Existing single Convolutional Neural Network (CNN) models lack feature variability and real-world generalization.

Purpose of the Study:

• To develop an enhanced deep learning framework for accurate coffee disease classification.
• To integrate complementary feature extraction from multiple CNNs with advanced feature selection.
• To improve computational efficiency and accuracy in coffee disease identification.

Main Methods:

• A hybrid deep learning framework integrating GoogLeNet and ResNet18 for feature extraction.
• Dimensionality reduction using Principal Component Analysis (PCA) and Singular Value Decomposition (SVD).
• Feature selection via Analysis of Variance (ANOVA) and Chi-square tests, trained with an Adam optimizer.

Main Results:

• Achieved 99.78% accuracy on the BRACOL dataset for coffee disease classification.
• Demonstrated precision, recall, and F1-scores exceeding 99% across all classes.
• Successfully integrated multiple DL architectures with feature selection for robust classification.

Conclusions:

• The proposed hybrid deep learning framework significantly enhances coffee disease classification accuracy.
• Systematic integration of diverse CNNs and feature selection methods addresses limitations of single-model approaches.
• This research provides a computationally efficient and highly accurate solution for sustainable coffee production.