手腕表面肌电信号的动作特征表示与识别

Action Feature Representation and Recognition of Wrist Surface Electromyogram Signals

为提高智能轮椅人机接口中表面肌电信号的正确识别率和识别效率,研究了基于小波包多尺度分解的特征表示和模式识别方法。首先把采集的表面肌电信号进行小波包分解,然后用小波包系数构造特征基向量。然后,根据小波包系数与表面肌电信号能量之间的内在联系重构了特征向量。最后用非线性自回归神经网络实现了双通道表面肌电信号四种不同动作模式的分类。实验结果表明,用小波包系数重构的特征基向量可作为表面肌电信号的动作特征,并能有效的简化分类器的结构。

In order to improve the correct recognition rate and recognition efficiency of the surface electromyogram signal （sEMG） of the intelligent wheelchair man-machine interface, feature representation and pattern recognition methods based on multi-scale decomposition of wavelet packet are studied. Firstly, the sEMG signal is decomposed by wavelet packet, and then the wavelet packet coefficients are used to construct the feature vector. Then, according to the inner link between wavelet packet coefficients and the multi-channel sEMG energy, the feature vector is reconstructed. Finally, the classification of four kinds of different action modes of the dual channel of multi-channel sEMG is realized by using the nonlinear auto regressive neural network. Experimental results show that the characteristic basis vector reconstructed by the wavelet packet coefficient can be used as the action characteristics of the multi-channel sEMG, and can effectively simplify the structure of the classifier.