用于装配动作识别的肌电信号特征优化选择方法  

Optimal selection method of electromyographic signal features for assembly gesture recognition

作　　者：刘永[1] 宁蕊 李言[1] 杨明顺[1] 高新勤[1] LIU Yong;NING Rui;LI Yan;YANG Mingshun;GAO Xinqin(Faculty of Mechanical and Precision Instrument Engineering,Xi’an University of Technology,Xi’an 710048,China)

机构地区：[1]西安理工大学机械与精密仪器工程学院,陕西西安710048

出　　处：《西安理工大学学报》2023年第4期513-520,共8页Journal of Xi'an University of Technology

基　　金：陕西省重点研发计划项目(2021SF-421,2021SF-422);陕西省现代装备绿色制造协同创新中心自主研发基金项目(102-451421003)。

摘　　要：在采用机器学习方法进行动作识别的研究中,识别的准确率很大程度上取决于输入数据的特征。针对基于表面肌电信号的作业动作识别,进行了特征分析与优化选择方法研究。在对采集的作业手臂肌电信号进行平滑处理的基础上,定义并提取了肌电信号时域、频域及时频域的15个特征量;将从8个通道肌电信号的每帧数据中计算获得的120个特征值用于手势姿态的表征,并进行了归一化处理;使用极限梯度提升(XGBoost)算法和单变量特征选择(UFS)算法分别从特征量和特征值两个角度对所得信号特征进行识别贡献度的分析。分析结果表明,两种方法均可大幅消减冗余特征,并且能有效提高最终的识别准确率,其中采用UFS算法选取的特征在识别速度和准确度上更具优势。In the research on gesture recognition by using machine learning methods,the recognition accuracy largely depends on the characteristics of the input data.A feature analysis and optimization selection method are proposed for operating gesture recognition with the surface EMG signal.Based on the arm EMG signal which was acquired and smoothed,15 feature parameters are defined and extracted in the time domain,frequency domain and time-frequency domain;120 feature values are calculated for each frame data of 8 channels EMG signal and normalized to characterize a certain gesture;the extreme gradient boosting(XGBoost)algorithm and the univariate feature selection(UFS)algorithm are used to analyze the recognition contribution degree of the features from the two perspectives of feature parameters and feature value.The analytical results show that the two methods can not only greatly reduce redundant features,but also effectively improve the final recognition accuracy.The features selected by the UFS algorithm have more advantages in recognition speed and accuracy.

关 键 词：动作识别 表面肌电信号 特征选择 极限梯度提升算法 单变量特征选择算法 

分 类 号：TN911.7[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]