Authors

Abstract

The quality of white tea buds is the basis of the quality of finished tea, and sorting white tea buds is a laborious, time-consuming, and key process in the tea-making process. For intelligent detection of white tea buds, this study established the YOLOv5+BiFPN model based on YOLOv5 by adding a Bidirectional Feature Pyramid Network (BiFPN) structure to the neck part. By comparing the YOLOv5 and YOLOv3 through the ablation experiment, it was found that the YOLOv5+BiFPN model could extract the fine features of white tea buds more effectively, and the detection average precision for one bud and one leaf was 98.7% and [email protected] was 96.85%. This study provides a method and means for white tea bud detection based on deep learning image detection, and provides an efficient, accurate, and intelligent bud detection model for high-quality white tea sorting.

Abstract in Chinese

白茶嫩芽品质是成茶品质的基础，白茶嫩芽分选是制茶工序中费工、费时且关键的工序。为了实现白茶嫩芽智能化检测任务，本文通过建立白茶鲜叶数据集，基于YOLOv5模型，在Neck加入双向特征金字塔网络（BiFPN）结构，得到了YOLOv5+BiFPN模型。通过烧蚀实验对比YOLOv5模型和YOLOv3模型，发现YOLOv5+BiFPN模型可以更有效的提取白茶嫩芽中的细小特征，对一芽一叶的检测精度达98.7%，[email protected]达96.85%。本研究为白茶嫩芽检测提供了一种基于深度学习图像检测的方法与手段，为名优白茶分选提供了一种高效、准确、智能化的嫩芽检测模型。