四种手指动作的肘臂表面肌电信号的模式识别算法

发布时间：2018-02-03 16:31

本文关键词： 表面肌电信号(sEMG) 人工神经网络 MYO 特征提取手指动作　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着科技的进步,手指的动作识别有许多应用,在人类交往中手部动作占有重要地位。手指的动作在人机交互中发展趋势上升,对手指动作的识别还可以帮助残疾人和老人实现人机的互动,而且通过利用信号进行神经元分析,对患者的病情进行诊断,制订康复计划等,使之成为热门的研究项目。手指识别在实际生活中具有广泛的应用前景,然而对于手指动作的识别研究较少,手指动作是手势动作的基本单元,对于手势动作的研究一般采用基于电极的肌电信号来对手势进行识别,因此本文提出利用肘臂表面肌电信号sEMG(Surface Electromyography,肌电信号的一种)来对大拇指与食指的点击,大拇指与中指的点击,大拇指与无名指的点击,大拇指与小拇指的点击,四种手指动作识别进行研究。实验通过对表面肌电信号的分析,利用MYO手臂环来对四种手指动作进行数据采集,替代了传统的电极深入肌肉的采集方式。通过EMGlab对MYO采集的数据进行活动段处理,对其信号进行1000Hz的高频滤波,峰值重读,然后再对信号进行MLT分解,得到每个频道的模板波形,会显示在模板面板内,最后对模板波形和整段的sEMG信号进行平均绝对值(MAV),方差(VAR)等五种特征提取,同时作为模式识别算法的输入参数。算法选择反向神经网络(BP)来对肘臂sEMG信号的四种手指动作进行分类,因为BP神经网络广泛应用于前人的研究中,具有高度的普适性,自适应性强和结构稳定。实验结果表明BP分类器具有较高的识别准确率,其对应目标手指动作识别率在90.35%。通过对四种手指动作的识别结果,设计了游戏信息控制的人机交互界面,对四种手指动作和手指动作组合在人机交互研究中提供可行的设计。
[Abstract]:With the development of science and technology, finger movement recognition has many applications, which plays an important role in human interaction. The recognition of finger movement can also help the disabled and the elderly to achieve human-computer interaction, and through the use of signals for neuronal analysis, diagnosis of the patient's condition, the development of rehabilitation plans and so on. Finger recognition has a wide application prospect in real life. However, there is little research on finger movement recognition. Finger action is the basic unit of gesture action. In general, the electromyography based on electrode is used to recognize the gesture. Therefore, this paper proposes the use of the elbow arm surface EMG signal sEMG(Surface myography, a kind of EMG signal, to click on the thumb and index finger. Thumb and middle finger click, thumb and ring finger click, thumb and small thumb click, four finger action recognition. The experiment through the surface EMG signal analysis. The MYO arm ring is used to collect the data of four finger movements, instead of the traditional way of collecting the electrode deep into the muscle. The data collected by MYO is processed by EMGlab. The signal is filtered with high frequency of 1000Hz, the peak value is read again, then the signal is decomposed by MLT, and the template waveform of each channel is obtained, which will be displayed in the template panel. Finally, the template waveform and the sEMG signal of the whole segment are extracted by the average absolute value of sEMG, variance and VAR. etc. At the same time, as the input parameter of the pattern recognition algorithm, the algorithm chooses the reverse neural network (BP) to classify the four finger movements of the elbow arm sEMG signal. Because BP neural network is widely used in previous research, it has high universality, strong adaptability and stable structure. The experimental results show that BP classifier has a high recognition accuracy. The recognition rate of the corresponding target finger movement is 90.35. The man-machine interactive interface of game information control is designed through the recognition results of four finger movements. The combination of four finger movements and finger movements provides a feasible design for human-computer interaction.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044;TP391.41

【参考文献】