Neural network classification of electromyogram data for the construction of a human/computer interface

It is proposed in this work that an alternative means of interaction with a computer is possible. This method uses electromyographic (EMG) signals generated from voluntary muscle activity. The EMG signals were recorded while the subject performed various facial gestures. No movement below the subject's neck was required, implying a potential for use when limited voluntary muscle actuation is possible. Data recorded during the EMG acquisition sessions were formulated into input vectors and used to train a neural network. The single-layer network with a non-linear activation function was trained using the standard backpropagation of errors method. Various techniques to optimize the learning and generalization capabilities of the network were employed. The results obtained show that a human/computer interface based on EMG signal classification is possible. Additional experimentation points to the potential for neural network EMG classification in other applications such as eye tracking. Recommendations for further study and implementation of an interface are also presented.