基于表面肌电信号的手腕动作模式识别.doc

基于表面肌电信号的手腕动作模式识别
张启忠*席旭刚马玉良罗志增佘青山
（杭州电子科技大学智能控制与机器人研究所，杭州310018）

摘要：基于表面肌电信号的肢体动作模式识别是假手仿生控制的基础。为提高动作模式识别率，论文从肌电信号的产生机理出发，选取分别表征肌电信号形态特征-细节复杂度和整体自似性的近似熵指标和分维数指标，作为模式识别的特征向量；同时提出了一种具有增量学习能力的K最近邻(KNN)模型增量学习算法作为模式识别的分类器。在对10位受试者手腕的四个精细动作腕伸、腕屈、腕内旋、腕外旋的识别实验中取得了92.5%以上的正确识别率。论文同时对增量学习能力对分类器动作模式识别率的影响做了对比实验，当假肢使用者生理变化时，以KNN模型增量学习算法作为分类器比采用不具增量学习能力的KNN模型算法的识别率高4.5%。上述实验表明，论文所提肌电信号动作模式识别方法方案合理，具有应用价值。
关键词：肌电信号；肌电假肢；KNN模型增量学习算法；近似熵；分维数
中图分类号：TP241.3文献标识码：A

Studyonwrist’smulti-movementpatternrecognitionbasedonsEMGdoi:10.3969/j.issn.0258-8021.2013.03.000
收稿日期：2012-00-00，录用日期：2013-00-00
基金项目：国家自然科学基金(61172134)；浙江省自然科学基金（Y1111189，LY12F03007）;浙江省科技计划项目(2012C33075)
*通信作者。E-mail:[e-mail]

ZHANGQi-ZhongXIXu-GangMAYu-LiangLUOZhi-ZengSHEQing-Shan
(InstituteofIntelligentControlandRobotics,HangzhouDianziUniversity,Hangzhou310018,China)
Abstract:ActionpatternrecognitionoflimbsusingsEMGisthebasisforbioniccontrolofaprosthetichand.InconsiderationofthegenerationmechanismofsEMG,theapproximateentropyandthefractaldimension,whichfeaturethesEMG’smorphologicalcharacteristicsincludingcomplexityandoverallself-similarity,waschosenasthefeaturevectorofpatternrecognitiontoimproveactionmoderecognitionrate.Inthemeantime,aKNearestNeighbor(KNN)modelincrementallearningmethodwithincrementallearningability,waspresentedasaclassifierofpatternrecognition.Inpatternrecognitionexperimentofclassifyingfourfinemovementsofthewrist(namelywristextension,wristflexion,wristpronation,wristsupination)with10participants,thecorrectmoderecognitionrateisabove92.5%.Inacontrastexperimentthatwasdesignedtoevaluatetheeffectsoftheincrementlearningabilitytotheactionmoderecognitionrate,thecorrectrecognitionrateis4.5percenthigherthanKNNmodearithmeticwithoutincrementallearningabilitywhentheprostheticuserschangedphysiologically.TheaboveexperimentalresultshowsactionmoderecognitionmethodbasedontheEMGisreasonableandpractical.
Keywords:electromyography;myoelectricprosthesis;KNNmodelincrementallearningalgorithm;ApproximateEntropy