The improvement of hand gesture recognition based on sEMG by moving average filtering for features

摘要

Surface electromyographic signal (sEMG) is a kind of bioelectrical signal, which records the data of muscle activity intensity. Hand gesture recognition based on sEMG plays an important role in the hand rehabilitation robotics. Improving the recognition accuracy of hand gesture is the goal always pursued by researchers. This study purposes to evaluate the performance of moving average filtering for features in improving the accuracy of hand gesture recognition based on sEMG. Firstly, empirical mode decomposition was applied on the effective motion segments extracted by sliding windows with different size and stride. Afterward, 21 features were extracted from original signal and each of the intrinsic mode function. Subsequently, moving average filtering with different orders were used to further suppress the noise in each feature series to improve the recognition performance. Finally, two representative machine learning methods (including decision tree and K nearest neighbor) were used to evaluate the performance. Compared with the original features, the intra-subject recognition accuracy rate after feature processing for KNN and DT increase by 17%, and 6%, respectively. The accuracy of the inter-subject across five subjects after feature processing increase by 28% (KNN), and 12% (DT), respectively. The above results prove that the moving average filtering for features could effectively improve the performance of hand gesture recognition.