An enhanced deep learning-based framework for diagnosing apple leaf diseases.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate apple leaf disease identification is crucial for crop production and agricultural sustainability.
• Existing methods may lack efficiency or require substantial computational resources for real-time application.

Purpose of the Study:

• To introduce E-YOLOv8, a lightweight and improved version of YOLOv8, for real-time apple leaf disease detection.
• To enhance detection accuracy, especially for small lesions, while minimizing computational cost.

Main Methods:

• Developed E-YOLOv8 by integrating GhostConv and C3 fusion for efficient feature extraction.
• Incorporated CBAM attention and a custom FPN for improved multi-scale feature fusion.
• Conducted large-scale evaluations on apple leaf disease datasets and tested on edge devices.

Main Results:

• E-YOLOv8 achieved 93.9mAP0.5 with only 5.3 GFLOPs and 1.8 million parameters.
• Demonstrated a 33.9x reduction in computational cost compared to YOLOv8l.
• Outperformed recent state-of-the-art detectors in experiments.

Conclusions:

• E-YOLOv8 offers a highly accurate and computationally efficient solution for apple leaf disease detection.
• The model is suitable for real-time implementation on resource-limited edge devices in practical agricultural settings.