The study on feature selection strategy in EMG signal recognition

In this paper, the wavelet package-based feature extraction method is utilized to get the discriminative feature set. Furthermore, two categories of feature selection mechanisms, i.e., filter-based and wrapper-based feature selection, are introduced. As the classic representatives, the rough set and principal component analysis techniques are subject to the filter-based feature selection mechanism while the sequential forward search and backward elimination techniques are subject to the wrapper-based mechanism. All the four techniques are employed to refine the full feature set and the comparison result is analyzed with the adoption of support-vector-machine classifier. Experiments show that considering the classification accuracy and time consumption, the wavelet package-rough set is an advisable feature extraction and reduction method for the electromyographic (EMG) signals.