支持向量机在表面肌电信号模式分类中的应用

采用小波变换的方法对实验采集的原始四通道表面肌电信号(sEMG)进行了分析,并提取小波分解系数的奇异值构建特征矢量,利用“一对一”分类策略和二叉树设计的多类支持向量机(SVM)分类器,很好地实现了对前臂8种运动表面肌电信号的模式分类,8种运动模式的平均识别率为98.75%.研究表明SVM分类准确率明显优于传统的BP神经网络、Elman神经网络和RBF神经网络分类器,且识别精度高,鲁棒性好,对肌电信号及其他非平稳生理电信号的模式识别,提供了一种具有良好应用前景的新方法.

Application of Support Vector Machine in Pattern Classification of Surface EMG

A new pattern recognition method of surface electromyography(sEMG) is proposed, based on wavelet transform and support vector machine(SVM).The original four-channel sEMG signals from four corresponding muscles are analyzed with wavelet transform,then the singular values of wavelet decomposition coefficients are extracted to be the signal characteristics to construct eigenvector.A new multi-class SVM classifier is designed according to the "one by one" classification strategy and binary tree.Experiment results show that the eight forearm movement patterns can be well recognized after training by the SVM with average recognition rate up to 98.75%,and that the SVM can sort out sEMG eight movement patterns more accurately than conventional BP neural,Elman neural and RBF neural networks,with high robustness also provided.The method proposed can be directly applied to the pattern recognition of other nonstationary bioelectric signals.