DeepEMPR: coffee leaf disease detection with deep learning and enhanced multivariance product representation.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant diseases significantly impact crop yields and agricultural sustainability.
• Accurate and rapid disease identification is essential for effective crop management.
• Artificial intelligence (AI) offers potential for automated disease detection systems.

Purpose of the Study:

• To enhance the classification of coffee leaf diseases and estimate their severity.
• To develop and evaluate a novel image preprocessing technique for improved disease detection.
• To assess the performance of various convolutional neural network (CNN) architectures for this task.

Main Methods:

• A novel preprocessing approach using enhanced multivariance product representation (EMPR) to decompose and reconstruct coffee leaf images.
• High-dimensional model representation (HDMR) was applied to enhance image contrast, highlighting diseased leaf areas.
• Evaluation of popular CNN models including AlexNet, VGG16, and ResNet50 for disease classification and severity estimation.

Main Results:

• The VGG16 model achieved the highest classification accuracy at approximately 96%.
• All evaluated CNN models demonstrated strong performance in predicting disease severity, with accuracies over 85%.
• The ResNet50 model notably surpassed 90% accuracy in severity prediction.

Conclusions:

• The proposed image preprocessing method significantly improves automated coffee leaf disease identification.
• CNN models, particularly VGG16 and ResNet50, are effective for accurate disease classification and severity assessment.
• This research advances automated crop health monitoring systems, contributing to sustainable agriculture.