Classification of Hand Motions from Surface Electromyography with Rough Entropy

It is critical to extract effective features from surface electromyography (SEMG) for further classification of multiple hand motions. Entropy-based approaches are effective in modeling the fuzziness and uncertainty of SEMG data, and various entropic measures have been studied in the past years. However, one of the major limitations of those proposed entropic measures is that the parameter selection is still empirical. Rough entropy (REn), which is defined based on the rough set theory, has shown good ability to process the fuzzy and uncertain data. In this paper, we proposed to apply the parameter-free REn to characterize SEMG signals. Further, to comprehensively evaluate the performance of REn feature, we compared it with four other popular entropies, namely, fuzzy entropy, permutation entropy, wavelet entropy and sample entropy. Different classifiers including support vector machine (SVM), AdaBoost and linear discriminant analysis were also adopted to evaluate the consistency of different entropies. Experimental results indicate that REn outperforms other entropies with significant improvement with all the three classifiers, and the best classification accuracy is achieved by REn with SVM (97.39+/-1.74%) in four channel combination. It suggests that the parameter-free REn has the potential to be used for real-time control of SEMG-based multifunctional prosthesis.