Detection of Apple Leaf Diseases Based on LightYOLO-AppleLeafDx

Area of Science:

• Agricultural technology
• Computer vision
• Plant pathology

Background:

• Early detection of apple leaf diseases is critical for effective orchard management and maximizing crop yields.
• Existing methods may lack the efficiency or accuracy needed for real-time field applications.
• Developing lightweight, high-performance models is essential for precision agriculture.

Purpose of the Study:

• To introduce LightYOLO-AppleLeafDx, an improved, lightweight YOLOv8-based framework for apple leaf disease detection.
• To enhance detection accuracy and recall while minimizing model size and computational load.
• To validate the model's practical applicability in real-time agricultural settings.

Main Methods:

• Developed a lightweight detection framework, LightYOLO-AppleLeafDx, by enhancing the YOLOv8 model.
• Integrated novel modules: Slim-Neck, SPD-Conv, and SAHead, to optimize network architecture.
• Conducted ablation studies to confirm the contribution of each module to performance.
• Evaluated the model on key metrics including precision, mAP, model size, and detection speed.

Main Results:

• LightYOLO-AppleLeafDx achieved high performance: 0.930 precision, 0.965 mAP@0.5, and 0.587 mAP@0.5:0.95.
• The model is significantly lightweight (5.2 MB) and supports fast real-time detection (107.2 FPS).
• Demonstrated practical deployment feasibility on RV1103 hardware with 14.8 FPS detection speed via NPU.

Conclusions:

• LightYOLO-AppleLeafDx offers an efficient and accurate solution for apple leaf disease identification.
• The lightweight design and real-time capabilities make it suitable for precision agriculture applications.
• This framework addresses the need for accessible, high-performance tools in modern farming.