Tea Bud Detection Model in a Real Picking Environment Based on an Improved YOLOv5

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Automated tea picking relies on accurate tea bud detection.
• Challenges include complex environments, small targets, and blurry focus.

Purpose of the Study:

• To propose a high-performance tea bud detection model for automated tea picking.
• To address limitations of existing models in detecting small and occluded tea buds in challenging conditions.

Main Methods:

• Developed an improved YOLOv5 model named YOLOv5-tea.
• Incorporated Simplified Spatial Pyramid Pooling Fast (SimSPPF) in the backbone.
• Utilized Bidirectional Feature Pyramid Network (BiFPN) and Omni-Dimensional Dynamic Convolution (ODConv) in the neck network.

Main Results:

• The improved model demonstrated a 4.4% increase in precision, 2.3% in recall, and 3.2% in mean average precision.
• Enhanced detection accuracy for small, occluded, and fuzzy tea buds.
• Achieved improved inference speed compared to the initial model.

Conclusions:

• The YOLOv5-tea model significantly improves tea bud detection performance.
• The model offers practical significance for automated tea harvesting in complex environments.