Classification of surface EMG signal based on wavelet transform with nonlinear scale

Surface EMG (sEMG) signal is a complex nonlinear, non-stationary signal. In this paper, wavelet transform with nonlinear scale (NWT) is introduced. Due to the gradual shortening of its time-resolution, NWT is good at extracting the precise time-frequency information from sEMG signal. First, every sEMG signal (30 sets are for forearm supination and 30 sets are for forearm pronation) is transformed into intensity distribution (time-frequency distribution) by NWT. And then the feature vector is determined from the characteristic roots which are obtained from the intensity distribution by principle component analysis. At last, the two patterns of sEMG signals are identified by BP neural network. The results show that the accurate classification rate is higher gained by NWT than by two conventional time-frequency distributions. At the same time, the calculating complexity of neural network is decreased greatly.