Research On Pattern Classification Method And Application Of Upper Limb Action Based On SEMG Signal

Surface electromyography(sEMG)signal was a kind of complex bio-electric signals,generated by human skin after myofibril contraction-It can reflect nerves and muscles’motion in a way and has great practical value in clinical medical nerves and rehabilitation medical field and so on.Introducing sEMG into the robotic system can understand the patient’s active motion intention and enhance autonomy of patient in rehabilitation training,research signal feature fusion and classification method is a very important content.Aiming at the characteristics of hemiplegia patients whose one-side limbs motion function is destroyed,combined with the needs of rehabilitation robot system,research on the recognition method of upper limb rehabilitation motion based on sEMG signal.The main work in this thesis was concluded as below.(1)The signal acquisition system is based on the characteristics and influencing factors of sEMG signal.The collected signal come pre-treated.The Butterworth digital filter is used to filter out the motion artifacts noise.The feature vector matrix constituted of the time domain and frequency domain characteristic values which are extracted from filtered signal.The matrix is the input signal of feature fusion.(2)In order to shorten the process of iteration and improve sparsity,adding L1/2 regularization constraint to NMF algorithm so that it can obtain more accurate results in feature fusion.Meanwhile,this method can enhance the similarity of features in the same action pattern and expand the difference among the characteristics of different action modes.It can lay a solid foundation for classification.(3)In the classifier design,Multilayer Extreme Learining Machine was used to map low dimension feature for raw data and improve the classification speed and the accuracy of identification,which parameters of each layer are trained by the encoder.The last layer of the classifier use TSVD method to solve M.P.generalized inverse matrix.It can solve the problem of unstable data which caused by man-made interference effectively during the data acquisition.It can also enhance the stability of the classifer.(4)Design of the rehabilitation action recognition system,which has the function of storing the signal and classifying the movement pattern.The human-computer interaction interface was designed by Matlab GUI modules.lt makes the whole system more perfect.Experiments of offline data simulation and online test showed a satisfied result of the algorithm proposed by this thesis.