Classification techniques for the extraction of nonperiodic transients from noise: an application in electromyography

A new set of techniques for the extraction and classification of transients in a general background of other signals and noise is described. An example of its use in electromyography is described. The automatic signal-processing method classifies nonoverlapping transients from different sources into groups of closely similar shapes using a pattern recognition method. Each transient is described by a set of features to form uncorrelated factors which are then used in an adaptive clustering technique to group together transients from the same source. In particular, motor unit action potentials in a myoelectric signal can be classified into their motor unit groups using this method.