基于深度学习技术的卷烟陈列智能化督导的研究与应用  

Research and application of intelligent supervision for cigarette display based on deep learning technology

作　　者：李焕森 覃忠达 谭春 林光耀 秦岑 赵贝贝 黄庆庆 LI Huansen;QIN Zhongda;TAN Chun;LIN Guangyao;QIN Chen;ZHAO Beibei;HUANG Qingqing(Guangxi Zhuang Autonomous Region Tobacco Company of CNTC,Nanning 530022,China;Baise Branch of Guangxi Zhuang Autonomous Region Tobacco Company,Baise 53309,Guangxi,China)

机构地区：[1]中国烟草总公司广西壮族自治区公司,南宁市530022 [2]广西壮族自治区烟草公司百色市公司,广西百色市533099

出　　处：《中国烟草学报》2024年第4期113-122,共10页Acta Tabacaria Sinica

基　　金：中国烟草总公司广西壮族自治区公司科技计划项目“‘终端之眼’:零售终端卷烟陈列督导智能化研究”(201945000024132)。

摘　　要：为解决烟草公司客户经理在终端卷烟陈列督导过程中耗时长、效率低等问题,论文基于深度学习技术设计了一种卷烟陈列智能化督导系统(Cigarette Display Intelligent Guidance System,CDIGS)。CDIGS系统采用一种基于特征匹配和分类算法级联的卷烟商品识别算法,对移动终端拍摄的卷烟陈列图像进行识别,自动获取零售户的卷烟陈列、警示标识粘贴信息以及卷烟陈列达标率、陈列规格数、重点品牌上柜率等相关数据,并通过与预设的卷烟陈列标准模型进行比对,将二次陈列建议实时反馈至零售户。以广西百色市516家卷烟零售门店为例,对CDIGS系统进行验证,结果表明:①图片翻拍识别模型,对翻拍照片识别率达到79%,可以有效地提升拍摄图像的真实性;②警示标识检测、卷烟商品检测、卷烟牌号识别模型的准确率均达到97%以上,能准确地提取图像中的卷烟信息;③经简单介绍,卷烟零售门店人员可快速掌握系统的使用,卷烟陈列达标率持续提升,为推广应用奠定基础。可见,CDIGS系统为卷烟陈列督导提供了一种新的方式,能有效地指导零售户按规则自主进行卷烟陈列,并为烟草公司快速获取到零售户相关营销数据提供了新的手段。To solve the problems of long time consumption and low efficiency in the supervision of terminal cigarette display by customer managers of the tobacco company,this paper designed an intelligent cigarette display supervision system(Cigarette Display Intelligent Guidance System,CDIGS)based on deep learning technology.The CDIGS system uses a cigarette product recognition algorithm based on feature matching and a cascade of classification algorithms.It recognizes cigarette display images taken by mobile terminals,automatically obtains the retailer's cigarette display,warning label placement information,and data related to cigarette display compliance rate,display specification count,and key brand shelving rate.By comparing with the preset standard model of cigarette display,the system provides real-time secondary display suggestions to retailers.Taking 516 cigarette retail stores in Baise City,Guangxi as an example,the CDIGS system was validated.The results show that:(1)The picture re-photography recognition model has an identification rate of 79%,effectively improving the authenticity of captured images;(2)The accuracy rates of the warning label detection,cigarette product detection,and cigarette brand recognition models all exceed 97%,accurately extracting cigarette information from images;(3)After a brief introduction,retail staff can quickly master the use of the system,and the compliance rate of cigarette display continues to improve,laying the foundation for widespread application.Thus,the CDIGS system provides a new way for cigarette display supervision,effectively guiding retailers to autonomously arrange cigarette displays according to rules and offering a new means for tobacco companies to quickly obtain relevant marketing data from retailers.

关 键 词：卷烟陈列 卷烟识别 警示标志检测 翻拍检测 深度学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TS452[农业科学—烟草工业]