Classification of individual finger motions hybridizing electromyogram in transient and converged states

To classify the five individual finger motions from an electromyogram (EMG) signal, a classification system that hybridizes EMG signals in both the transient and converged states of a motion is proposed. The classifications of finger motions are executed individually in each state by a well-established artificial neural network (ANN). Then, the outputs of the two classifiers are combined. The efficacy of the result is evaluated via a piano-tapping task, in which the subjects are instructed to tap a keyboard with each of their five fingers. We use this task to compare the proposed hybrid system and a conventional converged system that uses an EMG signal only in the converged state. For five of the six subjects, the accuracy ratio of finger motions was better in the proposed method: approximately 85% for each finger except the second. Further analysis suggests two remarkable advantages of the hybrid method: (1) the output of the ANN is more credible, and (2) finger motion in the transient state (i.e., the early phase) is more predictable.