Application of the Wavelet Transform for EMG M-Wave Pattern Recognition

Abstract

<p> An investigation as to the appropriateness of the wavelet transform for surface electromyography (EMG) M-wave pattern recognition is described. The M-waves are obtained by stimulating the median nerve at the wrist to activate the motor units. Surface electrodes and a graded stimulus amplitude are used. The resulting M-waves are classified using both wavelet vectors and the traditional power spectral coefficients as features sets in the pattern recognition scheme. A novel system was developed to obtain M-wave collections from subjects in the laboratory and to perform both real-time and offline analysis.</p> <p> The results obtained from the left and right thenar muscles of 4 healthy females and 2 healthy males are presented. These results are further analyzed offline to determine the effects of a changing discriminatory threshold for both wavelet and power spectral pattern recognition techniques. In addition, intra-class and inter-class Euclidean distances are shown for the set of unique M-waves derived from using the different feature sets. A time-invariant wavelet transform is implemented to improve classification by eliminating errors due to latency shifts.</p> <p> The results show that the number of unique M-waves obtained usmg wavelet features is less sensitive to a variation in discriminatory threshold. It may be concluded that a wavelet based feature set shows slight improvement in M-wave pattern classification. The time-invariant wavelet offers further accuracy.</p>