【作者】 何乐生；

【导师】 宋爱国；

【作者基本信息】 东南大学 ， 测试计量技术及仪器， 2006， 博士

【摘要】 基于生物电信号的人机交互技术是近年来人机交互技术和遥操作机器人研究的前沿和热点之一。本文以国家重大基础研究973项目、教育部留学回国人员基金项目为背景,旨在系统地研究肌电信号产生和传递的机理,信号的获取方法,信号特征值的提取方法,基于肌电信号的动作识别方法以及肌电信号人机交互技术应用于遥操作机器人领域的可行性。本文首先概括了基于生物电信号的人机交互技术产生和发展的趋势。总结了肌电信号的产生和传播机制。针对神经肌肉接头在肌腹处有聚集性的事实,提出了一种可以降低表面肌电信号非平稳性的电极设置方案。构建了生物电信号采集试验平台。包括便携式的肌电信号放大器和基于虚拟仪器的采集软件平台。提出了一种通过肌电信号捕捉运动起始时刻的方法。认为这种方法可以省去冲动到达肌小节后和肌小节内的ATP发生化学反应的时间,大幅度地降低人机接口的反应时间。阐述了基于精确动作起始时刻同步信号的动作识别方法。证明了精确起始时刻对于动作识别的重要性。提取运动起始时刻以后256ms信号的统计特征值和小波时频特征值的20个主成分作为动作识别BP网络的输入。使用BP网络的学习速度自适应法优化网络训练过程。试验表明,采用该算法对8种手部动作进行识别的正确识别率可以达到95%以上。分析了肌电信号识别方法应用于遥操作机器人领域的可行性和优缺点,提出了一种遥操作机器手的控制策略。论文最后提出了今后研究的方向。更多还原

【Abstract】 The human-computer interactive technology based on bioelectricity is currently the frontier of the teleoperation systems and human-computer interaction. This thesis is under the supply of“973”Key Research Program and fund for the return scholars from abroad provided by Ministry of Education. The content of the thesis include the mechanisms of myoelectric signal’s emergence and transmission, how to record myoelectric signal, how to extract the features, how to classify upper limbs action base on myoelectric signal and how to apply the technology in the field of teleoperation robots.At the beginning, the tendency of the human-computer interactive technology based on bioelectricity is epitomed.Then the thesis summarizes the mechanisms of myoelectric signal’s emergence and transmission. And points out the innervation zones are centralize in the bellies of muscle, so the electrode system should be put on one side of the bellies of muscle to in order decrease the non-stationary of the signal.The experimental system of electromyography (EMG) is constructed in chapter 3. The system includes portable amplifier of EMG and virtual instrument software DAS.A method is given to recognize the human-being upper limbs action’s start moment by EMG. Using the method to recognize the start moment of given movement increases the efficiency of man-computer interface greatly. Because it saves the time of action potential transfer in the muscle fiber and the time of chemical reaction between Ca2+ and ATP in the muscle cell.An actions pattern recognition method based on accurate synchronization of start time is expatiated in the thesis. The importance for actions recognition of the exact start time of sampling is proved then. The time-domain statistical and time-frequency wavelet features of the 256ms EMG signal after start time is extracted. The first 20 primary components of the features are put to the recognition backpropagation network. And variable learning rate BP method is used as the train’s method. The experiments indicate that the correct ratio could be 95% by using this method to recognize 8 type of actions.The strongpoint and shortcoming of using the EMG actions recognition method on the field of teleoperation robot are also analyzed. And a control strategy of robot hand is put forword.At last, the future’s research direction in this field is pointed.更多还原