YOLO-APLD: A Lightweight Apple Leaf Disease Detection Model Based on Multiscale Feature Fusion

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate apple leaf disease identification is crucial for effective pesticide application.
• Conventional deep learning models face challenges with large model sizes and varying disease scales.

Purpose of the Study:

• To develop a lightweight and accurate algorithm for detecting apple leaf diseases.
• To improve upon existing deep learning models for real-time disease monitoring in orchards.

Main Methods:

• Introduced YOLO-APLD, an enhanced YOLOv8n model incorporating an EP-C2f module for feature representation.
• Utilized Focal-SIoU loss for improved bounding box regression and classification, especially for challenging samples.
• Implemented a bidirectional feature pyramid network (BiFPN) and Slim-neck structure for efficient multi-scale feature fusion and reduced model complexity.

Main Results:

• YOLO-APLD achieved precision of 88.5%, recall of 84.3%, mAP of 88.5%, and F1-score of 86.4%.
• Demonstrated significant reductions in FLOPs (22.2%), parameters (23.3%), and model size (17.5%) compared to YOLOv8n.
• Achieved a high detection frame rate of 90.3 f/s on edge devices, confirming real-time capability.
• Showcased generalizability on grape and tomato datasets.

Conclusions:

• YOLO-APLD offers a practical and efficient solution for apple leaf disease detection.
• The model provides valuable technical support for precision agriculture and on-site disease monitoring.
• The enhanced detection performance and reduced computational cost make it suitable for real-world orchard applications.