Convolutional Neural Network Based Human Movement Recognition Using Surface Electromyography

Surface electromyography (sEMG) signals are physiological signals with non-stationary and non-gaussian characteristics. For different subjects, the statistical characteristics will be different. At the same time, as the number of movement increases, it becomes more and more difficult to classify them with high accuracy. In this paper, convolutional neural network and wavelet packet decomposition are proposed for the classification of 49 human movements in NinaPro database 2. The result shows that the average recognition accuracy for all the 49 movements is 90.77% across all the 40 subjects. Compared with the results in other papers with the same set of data, the accuracy has been improved by 12.06%. The results indicate that the proposed algorithms are very helpful for the recognition of sEMG based hand movements.