An EMG Gesture Recognition Method based on Multivariate Variational Mode Decomposition

Surface electromyography(sEMG) can reflect the state of muscle activity which has important application value in human-computer interaction, prosthetic control and clinical diagnosis. In this paper, a gesture recognition method based on convolutional neural network(CNN) as well as multivariate variational mode decomposition(MVMD) is proposed. Firstly, the MVMD is used to decompose the EMG signals into several levels to extract the spatiotemporal characteristics of the multi-channel EMG signals, and then the convolutional neural network(CNN) is trained by the spatiotemporal characteristics. As experimental results show, compared with the simply EMG images, the performance of the proposed model is improved by 16.77%, 18.70% and 13.60% respectively. Compared with multivariate empirical mode decomposition(MEMD), MVMD uses less components, which can significantly reduce the training cost of neural network.