Filter Group Based Feature Extraction of MES Controller

Effective feature extraction is vital to reliable classification. To improve the accuracy of transient myoelectric signal (MES) pattern classification, a group of filter based time-series representation is proposed. Twenty-five filters with the same bandwidth but different center frequency divide the signal frequency spectrum into 25 sub-bands. It is shown that the myoelectric signal during four contractions (elbow flexion, elbow extension, forearm pronation, and forearm supination) exhibits distinct features in these sub-bands. Test platform of multifunction MES controller is designed. Experiments show that the proposed method can classify the four contractions successfully