A combination of AR and neural network technique for EMG pattern identification

Abstract

The EMG data acquired during voluntary movement of the active muscles of the disabled may provide useful control commands and information in functional electrical stimulation or in artificial prosthesis provided that the raw EMG data are property processed and identified. This technique may be used by the patients to transfer commands to their paralyzed extremities or artificial limbs. Combination of autoregressive and neural network technique to identify various functional hand movements is proposed. Functional hand movements such as palmar flexion and dorsiflexion, wrist pronation and supination, wrist flexion and extension, are identified. A fourth order parametric model is employed to evaluate the set of coefficients. The coefficients are then used as input for the neural network to identify the functional movement. Experiment was done on three healthy individuals and the rate of identification is shown to be adequate to be used in the development of either neural prostheses or artificial limbs.