Research On Recognition Method Of Lower Limb Action Pattern Based On Surface EMG Signals

Surface electromyography(sEMG)is a kind of electrophysiological signal formed during the muscle activity.Nowadays,the decomposition of surface EMG and recognition of human action based on surface electromyography(s EMG)is a research focus.Through research,researchers strive to create a more natural,convenient,simple and effective human-computer interaction mode,which for artificial limb control,mobile equipment control,sports electronic products and other industries,has far-reaching significance.In the field of multifunctional artificial limb control,the recognition of surface EMG signal pattern is a basic problem.In this thesis,the method of pattern classification and the extraction of EMG signal characteristics are explored and analyzed.The EMG signal mainly comes from the surface electrode of four different muscle tissues of the lower limb and the experiment uses the optimized upgraded support vector machine to perform the pattern recognition of surface EMG signal under lower limb movement,which covers three parts: pretreatment,signal acquisition and lower limb action recognition.This thesis compares and analyzes the characteristics of the frequency domain,time domain and wavelet features,and then establishes the energy characteristics of the wavelet packet function after the dimension reduction processing,which is regarded as a feature vector.In pattern recognition,the least squares support vector machine and standard support vector machine are selected from many recognition classifiers,and then the acquired feature vectors are input in this classifier.After a series of analysis and processing,the lower limb movements of the surface EMG signal are recognized.After the comparison experiment,the training time and the recognition of the correct rate of the classifier are analyzed.Finally,it is found that the operation time is shorter and the recognition rate is more accurate when the four kinds of lower limbs motion recognition are performed based on the least square support vector machine model based on particle swarm optimization.