Temporal Feature Extraction for Improving Myoelectric based Recognition of Prosthetic Hand

摘要

The physically disabled people with upper limb amputation may get benefited from the improved control of prosthetic hand that combines individual and multiple finger control. sEMG signals recorded from muscles are mainly used to control these prosthetic hands. The data patterns of sEMG generated during muscle contraction while performing a different finger movements are utilized to generate the control commands required by such controllers. In sEMG based PR system, various features are extracted and fed to the classifier. However, the major drawback using existing time domain features is the poor recognition rate. This research aims at improving the classification accuracy of sEMG based multi-fingered prosthetic hand using two novel TD features when combined with the existing feature set. Three feature sets are evaluated in terms of classification accuracy. The proposed method is validated on sEMG signal recorded by two electrodes placed on the forearm for operating ten different finger movements. ULDA, the feature projection is employed to reduce the dimensionality of feature vector size. Three classifiers (SVM, KNN and LDA) are implemented to evaluate the classification accuracy. An average accuracy of 94% across all eight participants for ten different finger movements using only two channels sEMG signal proving the significance of the proposed scheme.