Classifying surface electromyography with thresholding wavelet network

This paper reports some experiments conducted to investigate the performance of thresholding wavelet network in classifying single channel surface electromyography recorded from flexor digitorum superficialis muscle. The technique is based on the enhanced classification ability of a novel wavelet network proposed by the authors. The network is the combination of wavelet transform, wavelet thresholding and artificial neural network. The network extracts and select time-scale features of input signals based on the features ability to identify each signal. Since the selection is irrespective to the energy magnitude of each features, the network is also sensitive to small features emersed in strong noise such as surface electromyography. The results were very promising with classification accuracy of greater than 85%.