http://www.jsjkx.comDOI:10.11896/jsjkx.220300278到稿日期:2022-03-30返修日期:2022-08-21基金项目:国家自然科学基金面上项目(61972092);郑州市协同创新重大专项(20XTZX06013);河南省高等学校重点科研项目计划(21A520043)ThisworkwassupportedbytheNationalNaturalScienceFoundationofChina(61972092),CollaborativeInnovationMajorProjectofZhengzhou(20XTZX06013)andResearchFoundationPlaninHigherEducationInstitutionsofHenanProvince(21A520043).通信作者:段鹏松(duanps@zzu.edu.cn)WiDoor:一种近距离非接触式身份识别方法曹晨阳杨晓东段鹏松郑州大学网络空间安全学院郑州450002(caochenyang3325@gmail.com)摘要基于Wi-Fi感知的非接触式身份识别技术快速发展,在智能安防、人机交互领域展现出较好的应用潜力。研究发现,现有的轻量级身份识别模型在近距离场景下,识别准确率会随信号收发端距离的缩短而大幅下降。为此,提出了一种基于Wi-Fi感知的近距离非接触式身份识别方法WiDoor。在数据采集阶段,该方法借助菲涅尔传播模型对接收端天线部署方案进行优化,并重构多天线步态信息以获得更完整的步态特征;在身份识别阶段,使用串联卷积模块和多分支多尺度卷积模块相结合的轻量级卷积模型,在降低计算复杂度的同时保持了较高的识别准确率。实验结果显示,WiDoor在收发端间距为1m的10人数据集上识别准确率高达99.1%,且内置模型的参数量仅为同精度模型的2%,相比同类方法具有明显的优势。关键词:Wi-Fi感知;近距离身份识别;天线部署优化;卷积神经网络中图法分类号TP391WiDoor:Close-rangeContactlessHumanIdentificationApproachCAOChenyang,YANGXiaodongandDUANPengsongSchoolofCyberScienceandEngineering,ZhengzhouUniversity,Zhengzhou450002,ChinaAbstractTherapiddevelopmentofcontactlessidentificationtechnologybasedonWi-Fisensinghasshownexcellentapplicationpotentialinthefieldsofintelligenthuman-computerinteractionandintelligentsecurity.However,ithasbeenfoundthatinnar-rowindoorscenarios,accuracyofexistinglightweightidentificationmodeldecreaseswiththeshorteningoftransceiverdistance.Tosolvetheaboveproblem,aclose-rangeandcontactlessidentificationmethod,WiDoor,isproposed.Duringthedataacquisitionstage,WiDooroptimizesantennadeploymentatreceivingendbasedonFresnelpropagationmodel,andreconstructsgaitinforma-...
