Discrimination Among Hand Motions Using Combination of Features

Abstract

Electromyography is commonly used in signal monitoring for the rehabilitation of prosthetic systems. The extraction and selection of the features is equally critical for monitoring and controlling a more precise prosthetic device. We aim to create an invariant feature set which would allow amputees to control their prosthetics intuitively and precisely, no matter at which limb position the movement starts. This study introduces a new different set of feature which is Logarithmic Band Power fused with Spectral Amplitude. LDA classifier was implemented to evaluate the performance of various combinations of feature sets involving both time and frequency domain. Classification performance of some comparable feature sets along with the proposed feature set is evaluated on sEMG data. Data of ten participants performing four different motion classes, at three different limb positions was extracted for training and testing. Results demonstrate that, relative to other feature sets, the proposed feature set achieves a substantial reduction in the classification error rate. Achieving a classification accuracy of 83% when averaged across all subjects and limb positions, the proposed method is comparable to the existing state of the art techniques.