基于卷积神经网络的表面肌电信号手势识别

Research on surface EMG signal gesture recognition based on convolution neural network

运用卷积神经网络原理,实现一维多通道的表面肌电信号的手势识别,避免了复杂的前期表面信号的预处理,以及手工特征提取阶段。文中分别采集右手的握拳、向左、向右和展拳4种手势的表面肌电信号。然后将采集的四种不同手势的肌电信号进行切割与标记,生成不同信号长度的八通道信号的训练集与测试集,运用卷积神经网络的原理,分别对其进行卷积、下采样。经过试验研究发现,运用卷积神经网络处理一维多通道表面肌电信号,从而实现手势识别的算法是可行的,并且能够得到较高的识别率。

In this paper,the convolution neural network principle is used to realize the gesture recognition of one-dimensional multi-channel surface EMG signals,which avoids the pretreatment of complex pre-surface signals and the manual feature extraction stage. This paper collects the right hand fist,the left,right and the right hand gesture four gestures of the surface EMG signal. Then the four different gestures of the EMG signal are cut and marked to generate different signal length of the eight-channel signal training set and test set. Using convolution neural network principle for convolution and down-sampling respectively. Experiments show that it is feasible to use the convolution neural network to solve the one-dimensional multi-channel surface EMG signal,and then to achieve gesture recognition,and this method can get higher recognition rate.