T-YOLO: a lightweight and efficient detection model for nutrient buds in complex tea-plantation environments

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Accurate detection of tea nutrient buds is crucial for yield prediction and field management.
• Challenges in tea plantations include complex environments and visual similarity between buds and leaves.

Purpose of the Study:

• To develop a lightweight and efficient model for accurate tea nutrient bud detection.
• To improve upon existing detection models for small targets in unstructured environments.

Main Methods:

• Introduced C2fG2 and DBS modules into the YOLOv5 backbone and neck.
• Pruned the head network for lightweighting and integrated a dynamic detection head.
• Utilized the Global Wheat Head Detection (GWHD) dataset for evaluation.

Main Results:

• T-YOLO achieved a mean average precision (mAP) of 84.1%.
• Reduced parameters by 47% and FLOPs by 65% compared to YOLOv5.
• Outperformed YOLOv8 by 7.5% in mAP.

Conclusions:

• T-YOLO effectively detects small tea nutrient buds, supporting smart tea garden management.
• The model's lightweight and efficient design offers a reference for detecting small targets in other crops.
• Demonstrated superior performance against mainstream detection models.