Wavelet De-Noising of Electromyography
Zhang Qingju, Luo Zhizeng · 2006
Electromyography (EMG) became noisy in the collection and transmission. To eliminate the noise, a novel threshold value method based on the wavelet de-noise was proposed. Firstly, the obtained EMG signal was decomposed by the wavelet transform. Then, the decomposed wavelet coefficients were analysed by the weighted average of traditional soft-threshold and hard-threshold. Finally, the wavelet coefficients were recovered by the wavelet reconstructed algorithm and got the de-noised EMG information. Lots of experiments have proved the method had good performance in removing noise, synchronously, the character information was remained. The method collected the merits of soft-threshold and hard-threshold and made a good base for the pattern recognition of EMG artificial limb. At the end of the paper, the RBF neural network classifier based on the power spectrum analysis was designed to validate the improving wavelet de-noising method. The accurate recognition-rate of four motions is increased to as high as more than 90%