Coffee Leaf Rust Disease Detection and Implementation of an Edge Device for Pruning Infected Leaves via Deep Learning...

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Global warming and climate change exacerbate coffee leaf rust (Hemileia vastatrix), a significant threat to coffee plantation productivity.
• Current pesticide application for coffee leaf rust is often inefficient; targeted pruning of infected leaves can enhance treatment efficacy.
• Accurate detection of coffee leaf rust is crucial for informed pruning decisions and effective disease management.

Purpose of the Study:

• To develop and evaluate deep learning models (YOLOv5 and YOLOv8) for automated detection of coffee leaf rust (CLR).
• To assess the performance of trained models in real-time detection using various camera inputs and edge devices.
• To integrate the CLR detection system with a pruning solution for Arabica coffee farms.

Main Methods:

• A dataset of 2024 coffee leaf images was collected using digital mirrorless and phone cameras, with rust-infected images labeled as 'CLR'.
• Images were used to train YOLOv5 and YOLOv8 convolutional neural network (CNN) algorithms.
• Model performance was evaluated using mean average precision (mAP) and recall, with testing on diverse datasets and real-time scenarios.

Main Results:

• YOLOv8 achieved a higher mAP (73.2%) compared to YOLOv5 (70.5%) in detecting coffee leaf rust.
• Real-time detection of CLR was successfully deployed on an edge device with high confidence.
• The developed system was integrated with a custom-designed pruning device for Arabica coffee farms.

Conclusions:

• YOLOv8 demonstrates superior performance for automated coffee leaf rust detection, outperforming YOLOv5.
• The AI-powered detection system offers a viable solution for real-time monitoring and targeted intervention in coffee plantations.
• Integration with pruning technology provides a pathway to more efficient and sustainable coffee farming practices.