A Detection Algorithm for Citrus Huanglongbing Disease Based on an Improved YOLOv8n.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Citrus Huanglongbing (HLB) severely impacts citrus production, necessitating accurate disease detection for effective orchard management.
• Existing object detection models struggle with low accuracy in detecting HLB due to environmental factors like variable lighting, leaf occlusion, small leaf size, and disease similarity.

Purpose of the Study:

• To develop an improved object detection model, YOLO-EAF, to enhance the accuracy of Citrus Huanglongbing detection in natural orchard environments.
• To address the limitations of current models in feature extraction, fusion, and regression precision for HLB detection.

Main Methods:

• Proposed YOLO-EAF model, an enhancement of YOLOv8n, incorporating an Efficient Multi-Scale Attention Module (EMA) for improved feature extraction.
• Integrated Adaptive Spatial Feature Fusion (ASFF) module to enhance multi-level feature fusion and model generalization.
• Utilized Focal and Efficient Intersection over Union (Focal-EIOU) as the loss function to accelerate convergence and improve regression accuracy.

Main Results:

• YOLO-EAF demonstrated an 8.4% increase in precision over YOLOv8n, achieving 82.7% on a custom citrus HLB dataset.
• The F1-score improved by 3.33% to 77.83%, and mAP (0.5) increased by 3.3% to 84.7%.
• The model showed enhanced regression precision and robustness.

Conclusions:

• YOLO-EAF offers a significant improvement in detecting Citrus Huanglongbing, outperforming standard models under complex field conditions.
• The proposed model provides a novel technical approach for smart monitoring and management of HLB in citrus orchards.