An enhanced lightweight model for apple leaf disease detection in complex orchard environments

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Automated apple leaf disease detection is vital for yield optimization and loss prevention.
• Challenges include environmental variations (light, shading) and overlapping disease spots, reducing detection accuracy.
• Existing models face high computational demands, hindering real-time mobile deployment.

Purpose of the Study:

• To develop an enhanced lightweight model (ELM-YOLOv8n) for accurate apple leaf disease detection, especially for small targets in complex environments.
• To reduce computational resource consumption for efficient real-time mobile deployment.
• To improve the model's robustness against environmental interference and enhance differentiation between similar diseases.

Main Methods:

• Integrated Fasternet Block into the backbone and neck networks to decrease model parameters and computational load.
• Incorporated Efficient Multi-Scale Attention (EMA) for enhanced feature extraction and anti-interference capabilities.
• Designed a detail-enhanced shared convolutional scaling detection head (DESCS-DH) for improved edge information capture and multi-scale detection.
• Utilized the NWD loss function for more accurate localization and identification of small disease targets.

Main Results:

• The ELM-YOLOv8n model achieved a 94.0% F1 score and 96.7% mAP50, outperforming the standard YOLOv8n.
• Reduced model parameter count by 44.8% and computational load by 39.5%.
• Demonstrated superior performance in detecting small disease targets and handling complex backgrounds.

Conclusions:

• ELM-YOLOv8n offers a highly accurate and efficient solution for automated apple leaf disease detection.
• The model's lightweight design makes it suitable for deployment on mobile devices.
• The enhancements effectively address challenges in complex environments and improve detection of subtle disease indicators.