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

Mengdao Yang,Weihao Yuan,Gaojian Xu

DOI: https://doi.org/10.1038/s41598-024-53498-y

IF: 4.6

2024-02-04

Scientific Reports

Abstract:Currently, 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.

multidisciplinary sciences

What problem does this paper attempt to address?

The paper mainly addresses the issue of accuracy in identifying and classifying tea buds. In the intelligent development of the tea industry, accurate identification of tea buds is an important step in determining tea quality and classification. The researchers established a tea bud dataset called TBD by capturing four different types of tea buds (Anji White Tea, Huangshan Variety, Longjing 43, and Nongkang Early). They compared seven mainstream object detection algorithms (YOLOv4, YOLOv5, YOLOX, YOLOv7, EfficientDet, Faster R-CNN, and CenterNet) in experimental tests and found that the YOLOX algorithm performed the best, achieving high accuracy, recall rate, and F1 score. Subsequently, they used the YOLOX algorithm and the dataset to build a tea bud classification model, which is capable of distinguishing these four types of tea buds effectively.

The challenges mentioned in the study include the low accuracy of deep learning-based object detection algorithms in identifying tea buds and the lack of research on tea bud classification, which is indispensable in practical production. The main content of the paper is to use seven object detection algorithms to establish a tea bud recognition model and compare them, and then select the best algorithm (YOLOX) to combine with the dataset to build a classification model. The experimental results show that the established classification model has good recognition performance for the four types of tea buds, providing theoretical support for mechanical intelligent tea picking. Future research will be devoted to improving the robustness of the model to meet the needs of the tea industry.