摘要
针对下肢肌电信号(EMG)的多运动模式分类问题,提出了一种基于小波支持向量机(WSVM)的多类识别方法.在小波框架理论和SVM核方法的基础上,构造基于二叉树结构的WSVM多类分类器,采用多尺度分析对下肢EMG进行消噪处理和特征提取,将特征向量输入WSVM多类分类器.以水平行走为例对支撑前期、支撑中期、支撑末期、摆动前期和摆动末期等5个细分运动模式进行分类,并与传统的神经网络和高斯核SVM分类器进行比较.实验结果验证了所提方法的有效性.
Aimed at the multi-motion pattern classification problem of lower limb EMG （electromyo- graphy）, a WSVM （wavelet support vector machine）-based multiclass recognition method was proposed. Firstly, the MWSVM （multiclass WSVM） was constructed using the binary tree structure, based on the wavelet frame and the kernel method of SVM. Secondly, the de-nosing process and fea- ture extraction were done by the multi-scale analysis for the lower limb EMG, and then the eigenvector is as the input of the MWSVM classifier. Finally, five subdividing patterns were identified in levelground walking, i.e. support prophase, support metaphase, support telophase, swing prophase and swing telophase, as compared with traditional neural networks and Gaussian kernel SVM （support vector machine）. The effectiveness of the proposed method is validated by experimental results.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2010年第10期75-79,共5页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家高技术研究发展计划资助项目(2008AA04212)
国家自然科学基金资助项目(60705010)
浙江省自然科学基金资助项目(Y1090761)
关键词
神经网络
肌电信号
小波变换
支持向量核
小波支持向量机
neural networks electromyography
wavelet transform
support vector kernel
wavelet support vector machine
