Research on a Few-shot Learning for Identifying Genuine and Fake Cigarette Label Paper
ZHANG Chao
ZHANG Ting
XU Weibin
WANG Shuming
ZOU Hua
YANG Yang
Abstract：This study aims to address the issues of low efficiency,high error rate in traditional identification methods,and ex-cessive dependence on sample size in deep learning methods for counterfeit cigarette detection in China's tobacco industry.A new deep learning-based method for authenticating cigarette trademark paper is proposed.By improving the Resnet network architecture and combining YOLO model,metric learning,and parameter-free classifier algorithms,a multi-level feature enhancement model is constructed,effectively reducing the demand for counterfeit cigarette samples while significantly improving the model's generaliza-tion ability for unknown counterfeit cigarettes.Experiments are conducted using small box trademark papers of various brands and specifications,and the results show that the method achieves an identification accuracy rate of up to 98％.This forms an efficient identification process of"first identify product specifications,second observe features,and third authenticate authenticity"demon-strating its feasibility and effectiveness in practical applications.
Keywords：deep learningYOLO algorithmmetric learning methodparameter-free classifier algorithmcigarette label paper authenticity identificationsensory identification methodimproved Resnet classification algorithm
Publication Date：2025-08-20
Online Publishing Date：2025-12-12(First online date of this platform, not the publication date of the document)
Pages：7( 2076-2082 )