Intelligent Recognition and Sorting Machine for Fresh Tea Leaves Based on Deep Learning Algorithms

Abstract

The inconsistency in the quality and morphology of fresh tea leaves poses a challenge to automated and standardized tea processing. To address this issue, an intelligent fresh tea leaf sorting system integrating deep learning-based visual perception, object detection, and automated manipulation is proposed. A YOLO-based model is developed to classify fresh tea leaves into four commercial grades, while a three-axis Cartesian motion platform and flexible pneumatic picking mechanism are employed for automatic sorting. Experimental results show that the recognition model achieves a mAP50 of 95.2%. The prototype processes 2.5 kg of fresh tea leaves within 20 min, achieving a throughput of 7.5 kg/h with a mechanical damage rate of 2.8%. The proposed system demonstrates the effectiveness of combining deep learning-based visual perception with automated execution, providing a practical approach for intelligent visual sorting and agricultural automation.