Research Of Lower Limb Movement Pattern Recognition Based On Sruface Electromyography

Lower limb paralysis caused by a stroke makes patients and families suffer thetremendous pain. With the increse in the number of patients, it also causes the social burden.Traditionally, physicians uses freehand to rehabilitation training, but patients prone toboredom. Lower limb rehabilitation robot can replace the physician to completerehabilition, can improve efficacy and reduce burden. In order to achieve self-trainingpatients, and reflect the patient’s movement intentions. In this paper, the human lower limbmotion pattern recognition of basic research to do the following:Firstly, the stroke pathogenesis is elabotated in detail and the the mechainsm ofsurface EMG and its application is describes. After analyzing the features of lower limbmuscles, the tibialis anterior and gastrocnemius medialis are selectd as the control source,each gait cycle is divided into four motion modes of support early, interim support, finalsupport and swing phase.Secondly, the pretreatment end the feature extraction of surface EMG is a veryimportant part. Wavelet packet μ rhythm threshold de-noising has solved the interference ofEMG inclusion of physiological noise, and the noise canceling effect is better than thetraditional soft and hard threshold methods. The paper use50%of the “overlappingwindow” method to data segments, after the time domain, frequency domain and timedomain-the joint analysis, the wavelet transform of time-frequency analysis is used tocompute wavelet features singular values of each section, and enter the classifier for patternrecognition after building feature coding.At last, the GA-Elman network is proposed in order to the four segments of the lowerlimb gait state motion pattern recognition. It is the use of genetic algorithm to aptinize theinitial weights and thresholds of Elman neural network to solve the defects of thetraditional neural network convergence is solve and prone to local minima problems, afternetwork training and identifying, the average recognition rate was85%, and superior toconventional BP and Elman networks significantly. Taking into account the four lower limbgait movement patterns can be mapped to hidden Markov mode (HMM) in four states, thisarticle attempts to recognize lower limb gait States by the signal classification method ofHMM, after Baum-Welch algorithm for the recaluation of traning of parameters of HMM and the Viterbi algorithm to identify, resulting average recognition rate was93.75%. Therecognition effect is better than GA-Elman neural network, and it can control theimplementation of lower extremity exoskeleton more accurately.