Improved discrete fourier transform based spectral feature for surface electromyogram signal classification

An improved discrete Fourier transform (iDFT) is presented in this study as a novel feature for surface electromyogram (sEMG) pattern classification. It employs the principle that the spectrum of sEMG signals changes regarding different motions. iDFT feature focuses on global information of local bands to increase the inter-class distance. The experiment results showed that iDFT feature had a better separability than two other spectral features, auto regression (AR) and Power spectral density (PSD), both on experienced and inexperienced subjects. The optimal bandwidth is between 30 and 50 Hz and influence of division methods is not significant. With the low computation cost and property of insensitivity to sampling frequency, our proposed method provides a competitive choice for prosthetic control.