基于关键点匹配的烟包标签错误检测算法  

Cigarette Label Error Detection Algorithm Based on Key Point Matching

作　　者：王家寿[1] 李达[1] 袁湘云[1] 葛文[1] 王保云 WANG Jia-shou;LI Da;YUAN Xiang-yun;GE Wen;WANG Bao-yun(Hongyunhonghe Group,Kunming 650231,China;School of Mathematics,Yunnan Normal University,Kunming 650500,China)

机构地区：[1]红云红河烟草(集团)有限责任公司,中国昆明650231 [2]云南师范大学数学学院,中国昆明650500

出　　处：《湖南师范大学自然科学学报》2021年第5期136-142,共7页Journal of Natural Science of Hunan Normal University

基　　金：红云红河烟草(集团)有限公司科技项目(HYHH2018XX03)。

摘　　要：如何自动、准确地识别烟包标签错误,是烟草生产流水线上存在的难题。针对这一问题,提出了基于关键点匹配的烟包标签参数计算与错误类型判断算法。首先基于SIFT进行关键点确定,并提取扩展特征;接着进行相似点簇聚类与对应关系评估;然后在此基础上计算标签参数;最后依据标签参数判断错误类型。算法对倾斜、折角、垂直位移和水平位移的识别率均高于99%,虚警率和漏警率均低于5%,能够准确识别烟包标签的各类错误。How to automatically and accurately identify cigarette package label errors is a difficult problem in tobacco production line.In order to solve this problem,in this work,an algorithm based on the key point matching is proposed to calculate the parameters of cigarette bag label and detect the error type.Firstly,key points are determined based on SIFT,and the extended features are extracted.Secondly,similarity clustering and corresponding relationship evaluations are carried out.Then,the tag parameters are calculated.Finally,error types are judged according to the tag parameters.Our results show that the recognition rates of tilt,angle,vertical displacement and horizontal displacement are all higher than 99%,and the false alarm rate and missing alarm rate are lower than 5%,confirming that our algorithm can accurately identify all kinds of cigarette package label errors.

关 键 词：烟包标签 错误识别 SIFT 关键点匹配 

分 类 号：P391[天文地球—地球物理学]