Research And Application Of Pu-erh Tea Face Recognition Traceability Technology Based On Deep Learning

Pu-erh tea is a type of specialty tea produced in Yunnan province,particularly in the Puer city and its surrounding areas.It is one of the leading tea varieties in terms of production in China.As people’s demand for health and wellness grows,the product quality of Pu-erh tea has received increasing attention.However,in the supply chain of Pu-erh tea,there are still illegal businesses and individuals who use inferior tea leaves to impersonate high-grade tea,resulting in the phenomenon of substandard quality and adulteration.Traditional traceability technology cannot completely eliminate the problem of food fraud in tea products,and an imperfect traceability system can easily cause mistrust between upstream and downstream in the entire supply chain.Meanwhile,the quality of tea products is closely related to the reputation of the enterprise,which affects the reputation of Pu-erh tea.Therefore,strengthening the authenticity of the source data of Pu-erh tea is an urgent issue that needs to be addressed at present.This article focuses on the above-mentioned problem and takes Pu-erh tea compressed tea cakes as the research object.By collecting the images of both sides of the compressed tea cakes,we propose a Pu-erh tea face recognition traceability technology based on deep learning.We developed a tea face feature extraction network model and combined it with blockchain technology to deploy and verify the model in web applications and wechat mini program.The specific work is as follows:(1)The Pu-erh tea face datasets were constructed.Based on the different fermentation processes,three types of Pu-erh tea face datasets were constructed in this study,including raw Pu-erh tea face dataset,ripe Pu-erh tea face dataset,and mixed tea face dataset.The dataset includes 200 pieces of raw Pu-erh tea and 200 pieces of ripe Pu-erh tea.After collecting images of both sides,the images were preprocessed and data augmentation was performed.The successful construction of the dataset provides data support for research.(2)A Tea face feature extraction model Tea Face Net based on Mobile Net V3 was proposed.This work focuses on the task of feature extraction of tea face,by introducing the ECA-Net attention mechanism module to enhance the performance of the lightweight network Mobile Net V3,adjusting the parameters of the entire network,and optimizing and evaluating the model using Triplet Loss and Softmax loss functions.The experiments show that the proposed Tea Face Net network has recognition accuracies of 97.58%,98.08%,and 98.20% on the datasets of raw tea face,ripe tea face,and mixed tea face,respectively.Compared with classical network models,the proposed model achieves the highest recognition accuracy on the three datasets with fewer parameters,which achieves the goal of accurately verifying tea face.(3)The Pu-erh tea face recognition credible traceability system was built.The system integrates the Tea Face Net tea face verification model,and adopts dual storage of open source alliance blockchain framework Hyperledger Fabric and My SQL database for data storage,realizing trustworthy identification and information traceability of tea products,laying the foundation for the construction of a Pu-erh tea anti-counterfeiting regulatory system.