SVM based simultaneous hand movements classification using sEMG signals

Prediction of motion volitions is a practical issue in control of artificial limbs. Four classifiers are investigated in this paper to discriminate simultaneous hand movements based on pattern recognition of surface electromyographic (sEMG) signals. A sEMG signal processing tube composed of feature extraction, feature reduction and movements classification is proposed for offline myoelectric pattern recognition. Previous research was mainly devoted to individual hand movements classification. In this paper, several common tools are used for definition of movements. The results show that Support Vector Machine (SVM) outperforms the other three classifiers on both accuracy and model-training time. The user-depend classification accuracy reaches as high as 92.25% while the accuracy of user-independent is about 80%. The proposed classification method is a promising candidate to be used in prosthetic control for a rehabilitation robot in the future.