基于机器视觉与深度学习的烟用商标纸质量评价方法研究

打开文本图片集

关键词：机器视觉；深度学习；YOLOv8；商标纸质量检测；相似度计算

中图分类号：TS736；TP27 文献标识码：A DOI：10.11980/j.issn.0254-508X.2025.11.02

Study on Quality EvaluationMethod of Cigarette Trademark Paper Based on Machine Vision and Deep Learning

WANG Lyu1CAO Changqing'SI Yong1SHEN Qiang2DING Ran1LIN Sen'SHEN Zhiyan1XUE Chen² XIA Zhicheng1CHEN Renyu1XU Yanmin1ZHANG Jun'PENG Yunfa²ZHAN Ying2* （1.ShanghaiTbacoGroupCo.，d.，hanghaio82;2.anghaihuangheyiElectonicTechnologyDevelopmentCo.d， Shanghai，） （ E-mail： Zhanying@shmicrovision.com）

Abstract：Tivestigatetealcatiofmachevisionthologintegatingdeplearinginevaluatingtequalityispetioultsof cigaretetrademarkpper，tisstudyproposedaomprhensieevaluationmethod.Ahigresolutionindustrialcameraustomzedligt sources，ndspecializedftwaresystemswereemploedtoonstructaaotateddatasetDyamicthresholdOBfeaturedtectioptimizedRANSACregstration，ndulti-andfusionstrategieswereadopted，fectivelyeliinatingstichingseams.TeiagsRof 38.9dBandSSMofO.94.Forfeaturerecognition，theYOLOv8modelwasenhancedbyintroducingaCBAMatentionmodule，combined with a ResNet-34 backbone network and FPN multi-scale feature fusion，achieving 99.4% mAP50， 99.6% recall，and 99.0% precision on thetestet.Adual-branchSiamesenetworkwasdesignedtocomputesimilaritybyfusingSFTdescriptorsanddepsemanticfeatures， achieving average recognition accuracies of 97.64% for small box trademark paper and 95.85% for carton trademark paper. Key words：machine vision；deep learning；YOLOv8;cigarete trademark paperquality inspection；similaritycomputation

在当今竞争激烈的市场环境中，商标作为品牌的核心标识，承载着企业的声誉、形象与消费者的认知。（剩余9984字）