Oak-YOLO: A high-performance detection model for automated Oak seed defect identification.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Oak seeds are vulnerable to pest infestations, impacting germination and growth due to high starch content.
• Accurate defect detection is crucial for ensuring the quality and viability of oak seeds in forestry.

Purpose of the Study:

• To develop an efficient and accurate defect detection system for oak seeds using a novel YOLO-based model.
• To enhance the YOLOv8 architecture for improved feature extraction and defect representation.

Main Methods:

• Developed Oak-YOLO, integrating EfficientViT backbone and Ghost-DynamicConv detection head into YOLOv8.
• Implemented WIoUv3 loss function for optimized bounding box regression of complex defects.
• Conducted experiments on single- and multi-object datasets, including validation with mobile-captured images.

Main Results:

• Oak-YOLO achieved 94.5% mAP50 and 95.3% F1-score on an oak-intensive dataset with 132.2 FPS inference speed.
• Demonstrated robustness across different devices, achieving high mAP50 scores on smartphone-captured images.
• Outperformed existing YOLO models in accuracy and computational efficiency for seed defect detection.

Conclusions:

• Oak-YOLO presents a practical and efficient solution for real-time oak seed quality inspection.
• The model's enhanced architecture and loss function improve the detection of small and irregular defects.
• This advancement holds significant potential for improving forestry management and seed viability.