论文总字数：25490字

摘 要

对表面肌电信号进行模式识别可用于分辨手势动作及肌肉状态，其广泛用于医疗诊断、假肢控制等领域，本学位论文主要研究如何从肌电信号中正确的提取出有代表性的特征，提高分类正确率。

本文首先对国内外基于肌电控制系统的研究现状进行了阐述，之后对整个毕设的总体设计进行了介绍。本研究使用了两个电极对手臂两块肌肉进行了肌电信号提取，共采集了十类动作的肌电信号，并介绍了肌电信号采集中需要注意的事项。对于信号中存在的噪声干扰及基线漂移现象，对原始肌电信号进行了预处理，之后分别对信号进行时域特征、小波特征及小波包特征提取，并对小波基的选择进行了分析。接着利用线性判别分析在多类分类中的扩展应用方法进行动作分类，并对所有特征进行了比较。结果表明，小波包分解获得的特征可分性最好，识别率远高于其他特征，但所需时间最长。最后讨论了实际应用时特征的选择，并用平滑处理提高了分类正确率，结果显示，最优特征的十类动作的离线数据识别率均在90%以上，平均识别率在95%以上。

关键词：基线漂移、小波变换、线性判别分析

Abstract

Pattern recognition of surface EMG signals can be used to distinguish gestures and muscle states. It is widely used in medical diagnosis, prosthetic control and other fields. This dissertation mainly studies how to extract the representative features from the EMG signal and improve the classification accuracy.

In this thesis ，the methods of controlling prosthetic limbs based on myoelectrical control at home and abroad are reviewed firstly, and then the overall design of the whole set-up is introduced. Firstly, two electrodes are used to extract myoelectric signals from two muscles of the arm. A total of ten types of electromyographic signals are collected, and the matters needing attention in the acquisition of myoelectric signals are introduced. For noise and baseline drift in the signal, the original EMG signal is preprocessed in this study, and then the time domain features, wavelet features, and wavelet packet features of the signal are extracted, and the choosing of wavelet basis is analysis. Then we use the expanding application methods of linear discriminant analysis in multi-category classification to classify the actions and compare all the features. The results show that the feature obtained by wavelet packet decomposition is the best from the rate of classification, and its classification rate is much higher than other features, but it drops dramatically through calculating run time. Finally, we discussed the selection of features in practical application, and improved the classification accuracy rate by smoothing. The results showed that the classification accuracy rate of the ten types of the optimal features was more than 90%, and the average classification rate was above 95%.

KEY WORDS: Baseline drift, wavelet transform, linear discriminant analysis

目录

摘要 I

Abstract II

第一章 绪论 1

1.1 研究背景及意义 1

1.2 基于表面肌电信号的控制系统研究现状 1

1.2.1 表面肌电信号的处理方法 1

1.2.2 国内外研究进展 3

1.3 本论文的结构 4

第二章 肌电信号的采集与预处理 5

2.1 实验设备 5

2.1.1 肌电电极 5

2.1.2 数据采集卡 5

2.2 手势及肌肉的选择 6

2.2.2 肌肉的选择 7

2.3 皮肤预处理与信号预处理 8

2.3.1 皮肤预处理及注意事项 8

2.3.2 信号预处理 8

第三章 特征提取与分类 13

3.1 时域特征提取 13

3.2 小波变换 13

3.2.1 小波基的选择 15

3.2.2 小波系数特征 18

3.3 小波包变换 19

3.4 结果与对比 20

第四章 分类器的改进及讨论 23

4.1 线性判别分析 23

4.2 线性判别分析在多类条件下的应用 24

4.2.1 一对多组合分类器 24

4.2.2 两两组合分类器 25

4.3 分类器的改进 25

第五章 总结及展望 29

致谢 30

参考文献 31

绪论

研究背景及意义

表面肌电信号是通过电极从肌肉表面记录下来的神经肌肉系统活动信号，能够准确地反映肌肉活动状态^[1]。目前表面肌电信号的研究和应用主要集中在以下几个方面：

（1）假肢控制：肌肉活动状态发生变化后会采集到不同的表面肌电信号，且反应在肌电信号的数值特征上，并对应着不同的肢体动作，因此可用于假肢的控制^[2]。

剩余内容已隐藏，请支付后下载全文，论文总字数：25490字