SEMG Signal Recognition Based on Wavelet Transform and SOFM Neural Network

In this paper, we use one channel to collect the surface EMG signals of these actions separately such as elbow flexion, elbow extension, forearm supination and forearm pronation. Whereas the advantage of wavelet transform that it has fine frequency resolution at low frequencies, we can get a 4-dimension characteristic vector which is made up of 3 maximum values of detail coefficients (coefficients in D6~D4 levels) and 1 maximum values of approximate coefficient by using sym8 wavelet to decompose EMG to 6 levels. We construct a SOFM neural network and adopt the 4-dimension characteristic vector as the network´s input vector to identify the sEMG. It shows good identification effects to identify the 4 movements above.