Abstract:
In practical Electromyography (EMG) control, it is necessary to select appropriate number of surface EMG channels with ideal classification performance. In this paper, the contribution of different muscles to motor tasks is taken as the optimization criterion for optimal channel selection, and a channel selection method based on muscle synergy(MS) is proposed. Firstly, the raw EMG signal is preprocessed to extract EMG features, and then non-negative matrix factorization (NMF) algorithm is used to extract the muscle synergy matrix for each gesture and converted. Secondly, the muscle weight coefficients of each gesture on each EMG channel are summed to obtain the importance coefficients of all EMG channels. Finally, the classification is carried out by support vector machine(SVM), random forest(RF) and K-nearest neighbor(KNN) classifier. The method is tested by using surface EMG recorded in DB5 subdatabase of Ninapro database. The test results show that when extracting 10 optimal channels, compared with the sequential forward selection(SFS), Markov random field(MRF) and Relief-F channel selection methods proposed in previous studies, the recognition accuracy of the EMG subsets determined by this method is similar to that obtained by MRF and Relief-F method, and is slightly lower than that obtained by SFS method, but the computational cost is lower than that of SFS, MRF and Relief-F methods.