Design and experiment of tea winnowing parameter control system based on YOLO-AE.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Manual tea winnowing lacks precision, leading to inconsistent quality.
• Current methods rely on subjective observation, impacting efficiency.
• Automating tea winnowing is crucial for quality and consistency.

Purpose of the Study:

• To develop an automated tea winnowing method using deep learning.
• To enhance the accuracy and efficiency of white tea processing.
• To provide theoretical and technical support for precision tea winnowing equipment.

Main Methods:

• Improved YOLOv11 model (YOLO-AE) with ACmix and EUCB for enhanced detection.
• Region segmentation and convolution neural network for real-time proportion analysis.
• Integration of winnowing theory with AI for parameter determination.

Main Results:

• The YOLO-AE model achieved 94% recognition accuracy on the verification set.
• Detection performance improved by 2.1% with a 40% reduction in inference time.
• The system demonstrated consistent identification accuracy across different tea proportions, with <3% difference in high-quality batches.

Conclusions:

• The proposed deep learning approach significantly enhances tea winnowing accuracy and efficiency.
• This method offers a reliable solution for real-time quality assessment in tea processing.
• The study provides foundational technology for developing advanced, automated tea winnowing systems.