Research On The Hand Gesture Recognition Algorithm Based On SEMG

With the increasing of intelligent products,the importance of gesture in human-computer interaction is increasingly prominent.Hand gesture recognition based on sEMG signal is not affected by the external conditions such as using scene,light,temperature and so on.It will be more and more used in human-computer interaction scene.At present,most of the commonly used gesture recognition methods based on sEMG signal extract single features for analysis,while sEMG signal has the characteristics of nonlinearity and instability.The above feature extraction methods are easy to cause the extraction of sEMG signal features is not accurate enough.Therefore,this paper proposes two algorithms for single channel sEMG signal and multi-channel sEMG signal,aiming at the problems of low accuracy of gesture recognition based on sEMG signal,single feature extraction,single channel analysis method for multi-channel sEMG signal and ignoring the correlation of information between channels,and proves the calculation by simulation experiments The validity of law.The main research work of this paper is as follows:1.For single channel surface EMG signal,the method of ensemble empirical mode decomposition?EEMD?is used to decompose the preprocessed surface EMG signal.The first three natural mode functions?IMF?are selected from the decomposed signals for further feature extraction,and the absolute mean value?MAV?,power spectrum mean value(P_MEAN)and basic scale are selected Degree entropy?BSE?is used for linear feature fusion and kernel limit learning machine?KELM?is used for classification.Compared with support vector machine?SVM?and naive Bayes classifier?NB?,the results show that KELM is superior to the other two classifiers in classification accuracy and classification time.The algorithm is applied to six kinds of daily gesture data in UCI data set for verification,with an average recognition accuracy of 95.45%,which is better than the recent literature results using the same data set.2.For the multi-channel sEMG signal,considering the coupling relationship between the channels,the multi-dimensional empirical mode decomposition?MEMD?is used to decompose the sEMG signal,and the permutation entropy?PE?,autoregressive model?AR coefficient?and absolute mean value?MAV?of the first three IMF are calculated as the eigenvectors.In this paper,kernel entropy component analysis?KECA?,principal component analysis?PCA?and kernel principal component analysis?KPCA?are used to reduce the dimension of the feature vector,and KELM is still used to classify the feature vector.The experiment shows that the dimension reduction effect of KECA is better.The algorithm is applied to NinaProDB2 to classify 50 gestures,and the average classification accuracy is 86.01%.Compared with other literatures using the same dataset recently,the algorithm proposed in this paper has better classification accuracy and robustness.To sum up,this paper designs two sets of algorithms for single channel and multi-channel sEMG signals respectively,which are verified by experiments on UCI public data set and NinaPro public data set respectively.The comparison between this algorithm and other documents in the same period proves that this algorithm has certain advantages in recognition accuracy,recognition time and robustness.This paper provides a practical solution for the practical application of gesture in human-computer interaction.