基于肌电信号的行为识别的研究

整理文档很辛苦,赏杯茶钱您下走!

免费阅读已结束,点击下载阅读编辑剩下 ...

阅读已结束,您可以下载文档离线阅读编辑

资源描述

广东商学院数学与计算科学学院基于肌电信号的行为识别的研究1本科毕业论文(设计)基于肌电信号的行为识别的研究院(系)数学与计算科学学院专业信息与计算科学班级09信息与计算科学一班提交日期2013年月日毕业论文(设计)成绩评定表广东商学院2013-JX16-广东商学院数学与计算科学学院基于肌电信号的行为识别的研究2毕业论文(设计)指导教师评语及成绩成绩指导教师签名年月日毕业论文(设计)复评教师评语及成绩成绩复评教师签名年月日毕业论文(设计)答辩评语及成绩成绩答辩委员会主席签名年月毕业论文(设计)总成绩(五级记分制)院(系)负责人签名年月日TITLE:RecognitionResearchBasedOnTheBehaviorofTheEMGSignal广东商学院数学与计算科学学院基于肌电信号的行为识别的研究3MAJOR:InformationAndComputingScience广东商学院数学与计算科学学院基于肌电信号的行为识别的研究4内容摘要本文对前臂肌肉群多个位置(包括肱桡肌、桡侧腕屈肌、桡侧腕长伸肌、尺侧腕伸肌、尺侧腕屈肌)进行表面肌电信号的采集,用RMS(均方差)与AR模型两种方法分别对采集的肌电信号进行特征分析和特征提取,再通过不同的分类方法(包括监督式学习、LDA分类算法)对现有的信号数据进行模式识别,区别屈肘、屈腕、屈指和前臂旋转等多种动作,最终使用MATLAB实验软件绘图直观表示各种分类方法的识别率,得到多种分类方法的优劣程度并最终得出结论从而将该项的研究成果拓展到人体其他肌群及相关假肢的控制中,另外在仿生控制人工动力假肢研究领域同样具有重要的意义。关键词:特征提取模式识别MATLAB表面肌电信号AbstractInthisresearch,IcollectthesurfaceEMGsinseveraloftheforearmmusclegroups(includingbrachioradialis、flexorcarpi、carpiradialislongus、extensorcarpiulnarisandflexorcarpiulnaris).WiththemethodssuchasRMSandARmodel,IextractthefeaturefromtheEMGswhichhasbeencollected.Thenbyusingclassificationmethods(SupervisedLearning:TheLDAalgorithm),IusetheexistingEMGsinordertodifferentiatebetweenelbow、wristflexion,、flexor、forearmrotationandsoon.Ialsousetwomoresignaloptimizationstoraiserecognitionrate.Ultimately,weuseMATLABtoplotpicturesandintuitivelytoshowtherecognitionrate.Finallythebestrecognitionrateis98.611%.广东商学院数学与计算科学学院基于肌电信号的行为识别的研究5Keywords:PatternrecognitionFeatureextractionSEMG目录1引言···································································11.1理论研究····························································11.2国内外文献综述与研究现状·······················································12表面肌电信号的采集··································································32.1肌电信号的模型说明································································32.2肌电信号的数据说明································································42.2.1肌电信号的拾取··································································42.2.2肌电信号的位置··································································52.2.3动作展示··································································63表面肌电信号的预处理·······························································84模式识别····································································94.1特征提取·································································94.1.1时域特征值:RMS均方根························································94.1.2频域特征值:AR自回归模型····················································94.2监督式学习····································································114.3绘图展示····································································13广东商学院数学与计算科学学院基于肌电信号的行为识别的研究64.3.1表面肌电信号··································································134.3.2RMS特征值··································································144.3.3AR模型特征··································································155信号优化处理····································································165.1MajorityVote与去噪函数······················································165.2绘图展示·······················································176结论····································································18参考文献························································19附录····························································21广东商学院数学与计算科学学院基于肌电信号的行为识别的研究11引言表面肌电信号(SurfaceElectromyography,SEMG)是人自主活动中肌肉表层多个运动单位所发出的电位序列最终在皮肤表面通过电极检测得到的时间与空间综合叠加的结果[10],是神经肌肉系统活动时伴随的生物电信号。由于它在一定程度上关联着肌肉的活动状态与功能状态,因此也能够反映一定的神经肌肉活动状况[2],故在肌电信号在临床医学(如神经肌肉疾病诊断)、康复医学(如肌肉功能评价)、人机工效学(如肌肉工作的工效学分析)、体育科学(如疲劳判定、运动技术合理性分析、肌纤维类型和无氧阈值的无损伤性预测)、仿生学(如人体假肢控制具)[13]等方面均有重要的利用价值。1.1理论研究近年来,基于表面肌电信号识别研究在医学生物领域的作用越发凸显,研究学者遍布全球,研究文章也层出不穷,学者在对表面肌电信号进行识别研究时采用同样的步骤,即肌电信号的采集、信号优化处理,特征分析及提取、模式识别,但最主要的是肌电信号分析、特征提取与模式识别两个方面。特征提取的目的在于通过研究表面肌电信号的时、频域特征与肌肉结构以及肌肉活动状态和功能状态之间的关联性,从而利用SEMG的变化有效反映肌肉的活动和功能,其研究分析主要集中在时域和频域分析。特征提取就所利用的理论方法而言,可分为五个方面:时域法、频域法、时域-频域法、高阶谱及混沌与分形等。特征提取是基础,分类是关键,分类器可以为肌电假肢提供更可靠的控制信号。用于表面肌电信号的模式分类方法很多,其中模糊分类器和神经网络分类器的应用最为广泛。1.2国内外文献综述与研究现状在对表面肌电信号进行识别研究时,国内外学者往往从单一的分析方法及单一的分类方法中得出识别率并将此应用在实际生活的各个领域中。学者研究的思路大都相似,不同之处在于以下两点:第一点是对采集的表面肌电信号进行特征提取的方法。在国外方面,[20]Disselhorst-Klug(2008)利用时域方法(平均值)提取出SEMG的特征值,用于研究SEMG与肌肉力之间的关系;[22]Reddy(2007)利用时域方法(均方根值RMS)提取出SEMG的特征值,用于研究SEMG和运动位移的关系,从而实现了手指和腕关节模型的控制;[21]Sbriccoli(2003)分别利用时域方法(均方根值RMS)和频域方法(中位频率MF)提取出SEMG的特征值,用于研究肱二头肌SEMG的幅值和频谱特征。广东商学院数学与计算科学学院基于肌电信号的行为识别的研究2而在国内方面,[4]罗志增,杨广映(2003)根据实际肌电信号的随机性特征,对其建立AR模型,利用AR模型特征、参数与肢体运动的确定性关系实现仿生控制;[15]吴冬梅,孙欣,张志成,杜志江(2010)在人体屈伸肘部的过程中,选取人体上肢检测表面肌电信号应用不同的方法(均方根值RMS、肌电值iMEG)对优化后的表面肌电信号进行了特征提取。[23]罗志增,严庭芳(2008)利用时-频域方法小波变换对SEMG进行特征提取,用于SEMG的模式分类和肌电假肢的控制。第二点是模式识别的方法。用于表面肌电信号的模式分类方法很多,如模糊分类器和神经网络分类器。模糊分类器已在自动控制、人工智能、图像识别、农作物选中、商品评价、化合物分类、地震、气象预报、灾情预报、经济学、社会学、语言学、管理科学及医学等诸多领域得到了广泛应用。在表面肌电信号信号识别方面,开始利用该分类器进行处理,如[26]E.Zahedi(1995)利用模糊K-均值策略进行了3个自由度的动作识别;[27]刘建成(1999)也利用模糊神经网络直接对残肢的EMG动作进行识别,虽识别率70%以上,但有更好的实际应用价值;后者神经网络分类器在给足数量训练样本的前提下,网络就可以通过学习获得对运动模式进行分类的能力,如[24]王人成(1998)利用该网络对屈腕、伸腕、向内旋腕和向外旋腕四种运动进行识别,其识别率都在95%以上;[25]WilliamPutnam(1993)分别利用单层感知器和多层感知器对屈臂和伸臂两动作进行识别,识别率均可达95%;[28]R.Knox(1

1 / 27
下载文档,编辑使用

©2015-2020 m.777doc.com 三七文档.

备案号:鲁ICP备2024069028号-1 客服联系 QQ:2149211541

×
保存成功