Research On The Movement Of Finger Rehabilitation Based On Multi-channel SEMG Signals

In the contemporary society,cases of upper limb paralysis caused by stroke or accidental injury are also common.As a rising research subject in medical field,the concept of active rehabilitation has been put forward in view of the above injuries,which is a method of rehabilitation training with the patient's own physical information as the control source and the external equipment as auxiliary.This method,which can effectively reflect the patient's own "training participation",has a better practical effect over passive rehabilitation method.A surface electromyogram signal is a weak biological electrical signal which is collected on the surface of the human body when the muscle is moving.It can,to a certain extent,reflect the functional state of human nerves and muscles.Surface EMG signal has the advantages of high instantaneity,good imitation and no trauma.Thus,it has become the most commonly used signal source in the field of bionic hand rehabilitation.At present,there are still some problems in the pattern recognition of sEMG signal,such as few varieties of movement recognition and the inadequacy of stability and instantaneity.Therefore,under the premise of ensuring high recognition rate,how to improve the stability and instantaneity constitutes the main content of this paper,which also has significant theoretical and practical value from the perspective of rehabilitation medicine.The main research contents are as follows:(1)Based on the characteristics of the 12-channel sEMG signal acquisition system developed independently,the forearm plate was produced by 3D electrode array scanning,3D modeling and 3D printing technology.While according to the anatomy of human forearm muscle and the type of finger movement,the layout of the electrode was determined that ensure the stability of the acquisition experiment.Then,a new data sample space was obtained by using the sliding window method to segment the sEMG signal.(2)The features of sEMG data samples were extracted by time domain analysis method,Auto-Regressive parameter model method and time-frequency domain analysis method,which included six movements--flexing the thumb,index finger,middle finger,ring finger,little finger and all the fingers,followed by respective validity evaluations of each feature space.Then,the three feature matrices were dimensionally reduced through the method of principal component analysis(PCA).With the redundant information being removed,a more efficient and effective feature matrix was obtained.(3)Based on the BP neural network theory and support vector machine theory,combining with the six movements respectively in sEMG signal time domain feature matrix,Auto-Regressive parameter matrix and wavelet packet characteristic coefficient of relative energy coefficient matrix,the model parameters of each classifier were determined.Thus the BP neural network and multi-classification SVM based on sEMG were designed.Then,a square-mediation method that extracts the envelope of multi-channel sEMG features was presented to create the teacher sample label.(4)After the training and testing of the classifiers and the comparison of pattern recognition effect of the time domain characteristic,Auto-Regressive coefficient feature and wavelet packet coefficient relative energy feature of the sEMG signal in BP neural network and multi-classification support vector machine,it was concluded that the combination of wavelet packet energy coefficient and multi-classification support vector machine was the best.In order to solve the problems in the field of pattern recognition,the surface electrode array was developed in this paper,which could ensure the stability of the acquisition experiment to a certain extent.Combined with the sliding window method and the square mediation method,the final pattern recognition accuracy rate reached 98.78%,which provides a good reference to the theoretical basis for rehabilitation equipment and intelligent prosthetic control.