Characterization of surface EMG with cumulative residual entropy

The cumulative residual entropy (CREn) is an alternative measure of uncertainty in a random variable. In this paper, we applied CREn as a feature extraction method to characterize six hand and wrist motions from four-channel surface electromyography (SEMG) signals. For comparison, fuzzy entropy, sample entropy and approximate entropy were also used to characterize the SEMG signals. The support vector machine (SVM) and linear discriminant analysis (LDA) were used to discriminate six hand and wrist motions in order to evaluate the performance of different entropies. The experimental results indicate that the CREn-based classification outperforms other entropy based methods with the best classification accuracy of is 97.17±1.97% by SVM and 93.56±4.13 by LDA. Furthermore, the computational complexity of CREn is lower than those of other entropies. It suggests that CREn has the potential to be applied as an effective feature extraction method in the control of SEMG-based multifunctional prosthesis.