收稿日期:2023-04-12基金项目:商洛学院科研基金项目(21SKY003);陕西省教育厅专项科研计划项目(22JK0365)作者简介:张乐,男,陕西山阳人,硕士,助教GestureRecognitionBasedonMEMSInertialSensorZHANGLe,CHENLe-Xiang(CollegeofElectronicInformationandElectricalEngineering,ShangluoUniversity,Shangluo726000,Shaanxi)Abstract:AnovelgesturerecognitionmodelarchitecturewasproposedbasedongatedrecurrentneuralnetworkfortheutilizationofMEMSinertialsensorswithleveragingsmartphonesascarriersequippedwithbuilt-inMEMSinertialsensors,gesturemotiondatawasrequiredtoconstructtheLSTM-Dmodelbasedonthelongshort-termmemory(LSTM)network,aswellastheGRU-Dmodelbasedongatedrecurrentunit(GRU)network.Bothmodelsdemonstrateeffectivegesturerecognitioninthree-dimensionalspace.Ourevaluationonaself-builtdatasetshowcasesdesirableclassificationperformance,withtheLSTM-DandGRU-Dmodelsachievingaccuraciesof81%and85%,respectively.In-depthanalysisrevealsthattheGRU-DmodeloutperformstheLSTM-Dmodel,exhibitingreducedparameterization,shortertrainingtime,fasterandmoreaccuraterecognition,andenhancedstability.ThesefindingsprovidevaluableinsightsforadvancingthefieldofgesturerecognitionresearchemployingMEMSinertialsensors.Keywords:inertialsensor;deeplearning;recurrentneuralnetwork;gesturerecognitiondoi:10.13440/j.slxy.1674-0033.2023.04.006第37卷第4期2023年8月商洛学院学报JournalofShangluoUniversityVol.37No.4Aug.2023基于MEMS惯性传感器的手势识别张乐,陈乐翔(商洛学院电子信息与电气工程学院,陕西商洛726000)������������摘要:针对基于MEMS惯性传感器的手势识别问题,提出了一种基于门控循环网络的手势识别模型架构。以智能手机为载体,通过其内置的MEMS惯性传感器获取手势运动数据,构建了基于LSTM网络的LSTM-D模型和基于GRU网络的GRU-D模型,实现了在三维空间中的手势识别。提出的两种模型均有较好的分类效果,在自建数据集上,LSTM-D模型和GRU-D模型分别可获取81%和85%的准确率,综合分析发现GRU-D模型参数更少,训练时间更短,模型识别更快更准确,模型的稳定性更高,为基于MEMS惯性传感器的手势识别研究提供了一定的参考价...