基于支持向量机的手臂动作表面肌电信号模式分类方法研究

发布时间：2018-01-19 04:25

本文关键词： 表面肌电信号小波包变换支持向量机模式识别　出处：《吉林大学》2014年硕士论文　论文类型：学位论文

【摘要】：手臂动作表面肌电信号是利用表面电极从手臂皮肤表面记录下来的肌肉电信号，它可以量化的反映手臂进行动作时神经以及肌肉的功能状态。由于表面肌电信号的提取方式具有方便、准确、无创伤等优点，在康复医学、运动医学及智能机器人等领域都有广泛的研究与应用。随着信息科学技术的不断发展，准确的从表面肌电信号中提取有效特征，依据信号特征实现高分辨的动作模式识别，成为肌电信号控制仿生假肢技术的关键所在。本文依托吉林省科技发展计划重点项目具有温度、触滑觉临场感的仿生手臂研制与开发(批准号：20090350)，实施对手臂动作表面肌电信号模式分类方法的研究，以促进项目中肌电控制仿生假肢的实用化，因此本文的研究具有重要的科研价值与社会意义。本文的主要工作有： 1.明确表面肌电信号特点，结合局部解剖学相关知识，明确对手臂进行动作时贡献最大的两块肌肉，并确定电极采取信号的有利位置，利用贴片电极与肌电信号采集仪器完成了手臂常见动作模式的表面肌电信号的提取工作。 2.利用时域分析法、频域分析法及时频域分析法对所采集到的表面肌电信号进行了特征提取工作，分析数据结果，认为时域分析法与频域分析法具有片面性，确定由典型的时频域分析法的小波包方法来提取表面肌电信号的特征，最终由小波包系数的方差与能量作为特征元素组成特征向量。 3.分析了模式识别的主要方法，确定由标准支持向量机，最小二乘支持向量机，BP神经网络算法对所得到的特征向量进行模式识别。进行了参数选择方法的讨论与实验，分析了三种参数寻优方法的特点。 4.对三种模式识别算法的正确识别率及训练时间进行了统计分析，确定了应用粒子群方法进行参数寻优，，最小二乘支持向量机进行模式识别的优越性，其具有较高的识别率和较短的运算时间。虽然参数寻优的运算时间较长，但是分类器模型的训练与测试可分为两步，所以将参数寻优放入分类器模型的训练步骤中，利用获得的分类器模型进行动作模式识别的时间就可以大大缩短。 5.利用Matlab的GUI模块对手臂动作模式离线识别系统的开发，对表面肌电信号的各个处理环节进行整合，使系统变得可视化、易操作。
[Abstract]:The electromyography (EMG) signal of the arm movement surface is recorded from the skin surface of the arm using the surface electrode. It can quantificationally reflect the functional state of the nerve and muscle when the arm is moving. Because of the convenience, accuracy and no trauma of the surface EMG signal extraction, it is in rehabilitation medicine. Sports medicine and intelligent robot are widely studied and applied. With the development of information science and technology, accurate extraction of effective features from surface EMG signals. High resolution motion pattern recognition based on signal features has become the key of EMG control bionic prosthesis technology. This paper relies on the key project of Jilin province science and technology development plan has temperature. The research and development of the biomimetic arm of touch and slip telepresence (approval No.: 20090350) is carried out to study the classification method of electromyography (EMG) signals on the surface of the arm. In order to promote the utility of electromyoelectric control of biomimetic prosthesis in the project, the research of this paper has important scientific research value and social significance. The main work of this paper is as follows: 1. Make clear the characteristics of surface EMG signal, combined with the relevant knowledge of local anatomy, identify the two muscles that contribute the most to the arm movement, and determine the favorable position of the electrode signal. The surface electromyography (EMG) signal extraction of the common arm action mode is accomplished by using patch electrode and EMG signal acquisition instrument. 2. The time domain analysis and the frequency domain analysis are used to extract the features of the collected surface EMG signals, and the results are analyzed. The time-domain analysis and frequency-domain analysis are considered to be one-sidedness. The wavelet packet method of time-frequency domain analysis is used to extract the features of surface EMG signal. Finally, the eigenvector is composed of the variance and energy of wavelet packet coefficients as characteristic elements. 3. The main methods of pattern recognition are analyzed and determined by standard support vector machine and least square support vector machine. BP neural network algorithm for pattern recognition of the obtained eigenvector, parameter selection methods are discussed and experimental, and the characteristics of three parameter optimization methods are analyzed. 4. The correct recognition rate and training time of the three pattern recognition algorithms are statistically analyzed, and the superiority of using particle swarm optimization method and least square support vector machine for pattern recognition is determined. Although the operation time of parameter optimization is longer, the training and testing of classifier model can be divided into two steps. So the time of action pattern recognition using the obtained classifier model can be greatly shortened by putting the parameter optimization into the training step of the classifier model. 5. Using the GUI module of Matlab to develop the off-line recognition system of arm action pattern, and integrate all the processing links of the surface EMG signal, so that the system becomes visual and easy to operate.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】