YOLOX target detection model can identify and classify several types of tea buds with similar characteristics

Abstract

AbstractCurrently, the accuracy of tea bud identification is crucial in the intelligent development of the tea industry, and this is due to the fact that identifying tea buds is a key step in determining the quality of tea and distinguishing the categories. In this experiment, 3728 tea shoots with similar characteristics in four categories (Anji White Tea, Huangshan Seed, Longjing 43, and NongKang Early) were photographed to establish the dataset TBD (Tea Bud Dataset). In this experiment, we constructed a tea shoot recognition model. We used seven mainstream algorithms (YOLOv4, YOLOv5, YOLOX, YOLOv7, EfficientDet, Faster R-CNN and CenterNet) to conduct shoot recognition comparison experiments and found that the YOLOX algorithm performs the best with its Precision, Recall, F1 score, mAP 89.34%, 93.56%, 0.91, and 95.47%, respectively. Then the YOLOX algorithm combined with the dataset to construct the shoot recognition model, the shoots of four kinds of tea to establish a tea shoot classification model, the model to identify the Anji white tea shoots of Precision 76.19%, the yellow mountain species of Precision 90.54%, Longjing 43 Precision 80%, NongKang early to the morning of the Precision was 77.78%. The results of this experiment show that the established tea shoot classification model has achieved a better classification of the above four types of tea shoots, which can also understand the feasibility of mechanical intelligent tea picking and provide some theoretical support for the application of mechanical intelligent tea picking in practice.