Channel selection in multi-channel surface electromyogram based hand activity classifier

Abstract

Surface electromyogram (sEMG) is being extensively studied for development of limb prosthesis and for designing human-machine interfaces. To achieve acceptable accuracy in classification of hand activities using signals acquired from muti-channel sEMG system, the utility of various features and algorithms for reducing the size of the feature set have been reported extensively in literature. In this paper, an approach to select the channels that may be better suited for classification is proposed. The selection of channels is based on the feature selection algorithm namely minimal redundancy maximal relevance method. The proposed algorithm is applied on actual sEMG signals for two set of hand activities to assess the utility of the channels in activity classification. The results are validated by determining the extent of degradation in classification accuracies provided by the support vector machine classifier, when certain sEMG channels are excluded from classification.