Limb-function discrimination using EMG signals by neural network and application to prosthetic forearm control

The authors propose a method to estimate the motion intended by an amputee from his EMG (electromyographic) signals using a backpropagation neural network. The proposed method can discriminate the amputee´s intended motion among six kinds of limb-functions from multichannel EMG signals preprocessed by bandpass and smoothing filters. The cross-information among the EMG signals can be utilized to make the electrode locations flexible, and the bandpass filters can provide the amplitude and frequency characteristics of the EMG signals. Experiments on three subjects and four electrode locations demonstrate that the method can discriminate six motions of the forearm and hand from unlearned EMG signals with an accuracy above 90%, and can be adapted to some dynamic variations of the EMG signals by backpropagation learning