Using Wavelet Eigenvalues as the Input of Pi-sigma Network to Recognize Movement Pattern of Electromyography

In this paper, the noise mixed in the electromyography (EMG) signal is removed by using the wavelet transform firstly. Then the Pi-Sigma neural network uses the maximum of the wavelet coefficient under each scale as its input to recognize the movement pattern of EMG. At the same time, based on energy characteristic of the EMG we obtain the time of start and stop of the motion which used to control the movement of powered prostheses with the result of pattern recognition. The experiment indicates that the method of second wavelet de-noise can eliminate
the noise of EMG effectively. The real time control of powered prostheses can be realized base on the time information of pattern recognition under the condition of getting the correct result.