Welcome to Journal of Tea Science,Today is

Journal of Tea Science ›› 2021, Vol. 41 ›› Issue (2): 261-271.doi: 10.13305/j.cnki.jts.2021.02.008

• Research Paper • Previous Articles     Next Articles

Construction of Green Tea Recognition Model Based on ResNet Convolutional Neural Network

ZHANG Yi1, ZHAO Zhumeng2, WANG Xiaochang2, FENG Haiqiang3, LIN Jie1,*   

  1. 1. The Key Laboratory for Quality Improvement of Agricultural Product of Zhejiang Province, Zhejiang A&F University, Hangzhou 311300, China;;
    2. Institute of Tea Science, Zhejiang University, Hangzhou 310058, China;
    3. Planting Administration Bureau of Zhejiang Province, Hangzhou 310020, China
  • Received:2020-09-02 Revised:2020-10-26 Online:2021-04-15 Published:2021-04-13

Abstract: Green tea is the tea with the largest variety and output in China. Its appearance is an important basis for its classification. Image classification is one of the core technologies of computer vision, but its application in tea field is almost blank. Tea recognition still relies on the sensory evaluation methods by experts. This study collected 1713 pictures of 8 kinds of green tea (Lishui Xiangcha, Xinyang Maojian, Lu'an Guapian, Taiping Houkui, Anji Baicha, Biluochun, Zhuyeqing and Longjing). Based on the convolutional neural network, we explored the effects of ResNet model depth from the perspectives of model convergence speed, size, efficiency and identification balance. Finally, the ResNet-18 and SGD optimization algorithms were selected and a deep learning model was established to distinguish 8 kinds of green tea. The accuracy reached 90.99%, the recognition time of the single picture was only 0.098 s, and the model size was 43.7 MB. This paper provided the foundation for constructing a tea visual recognition model and applying it to the mobile terminals and provided a new accurate and efficient method for tea recognition.

Key words: convolutional neural network, deep learning, classification of green tea

CLC Number: