An application of Walsh functions for EMG classification

A pattern classification system that assumes no a priori information about the muscle state and treats the EMG (electromyogram) Walsh spectrum as a feature is presented. Walsh transformation and signal representation are used to reduce EMG signal dimensionality. The Walsh spectrum of EMG from a particular class forms a single cluster in the feature space. Thus, a one-dimensional representation of multidimensional data is obtained. Fourier and Walsh transformations for signal classification are compared. The Walsh transform estimation of the EMG data spectrum is faster and more unlimited with the bandlimited signals than the Fourier transform. The solution of the eigenvalue problem of the estimated EMG features by the Walsh transform gives greater separability of the EMG classes than the Fourier transform