| Surface electromyograph signal (sEMG) is potential signal collected on thehuman skin surface through the body surface electrode and produced by humanmuscle movement. Different hands grasping actions triggered different muscle groupactions, thus producing different electrical signals. sEMG has unique advantages inpattern recognition of human hands crawling actions, therefore, it is widely used inprosthetic control, rehabilitation, clinical medicine, sports science and many otherfields. For the use of sEMG to identify hand motion modes, it is critical of sEMGfeature extraction and pattern classification.The identification of manpower crawling action is based on the processing ofhuman forearm sEMG signal, thus, achieving the control of myoelectric prostheses forrehabilitation therapy for patients with disabilities. This paper studies thepre-treatment of sEMG feature extraction and pattern recognition. The main work isas follows:(1)Acquire sEMG, find the relationship between sEMG and muscle based onknowledge of human anatomy, sports medicine, etc. and determine the best location ofsEMG electrodes. sEMG signal collection instrument produced by the U.S. DELSYSis used to collect eight kinds of common hands grasping actions.(2)Pretreatment of sEMG, including segment detection filtering and activities.The main bands for sEMG is in20-500HZ. Therefore, using20-500HZ Chebyshevband-pass filter used de-noising, then using Chebyshev band-stop filter used forfrequency interference filter of50HZ. In this paper, the steady-state operation ofsEMG pattern recognition for staffing crawl. Using the moving average method toprocess instantaneous energy of sEMG sequence, and testing activities segment combined with threshold comparison method.(3)Feature extraction of sEMG, firstly common feature recognition methods,such as, time-domain characteristics, frequency-domain characteristics andtime-frequency domain characteristics, were analyzed. On this basis，we proposedfeature extraction methods combining wavelet packet energy spectrum and principalcomponent analysis.(4)Pattern classification of sEMG, firstly comparatively analyze Bayesianstatistical methods (Bayes) decision classifier, tfizzy classifier, neural networkclassifiers and support vector machine classifier based on statistical approach,combining with sEMG characteristics and the purpose of the study. Since SVMalgorithm has advantages of relatively complete theory, strong adaptability, globaloptimization, short training time, good generalization performance, complexity ofalgorithm being independent of characteristic sample dimension, and better robustness,etc. the paper selects SVM classifier for pattern recognition of staiffng crawlingactions.(5)The results of experiments and analysis. Firstly, select eight kinds of commonhand motion modes and four forearm muscles as source of sEMG. Experiments whichbased on time-domain feature extraction methods and time-frequency domain featureextraction methods used hand motion modes recognition rate, time-frequency domainfeature extraction methods is superior combination of time-domain characteristics.Study on different types of hand motion modes, and experiments results show that therecognition rate of common five kinds of hand motion modes averages100%, theaverage recognition rates of six and seven kinds of hand motion modes are also morethan97%, the average recognition rate of eight kinds of hand motion modes is90.5%.Thus, the experiments results have met the demands of sEMG prosthetic hand control.Finally, experiments used three surface electromyography electrodes, and the methodof pattern recognition for sEMG signal which based on the combination of waveletpacket energy spectrum and principal component analysis, for the study of differenttypes of hand motion modes. Experiments results show that the recognition rate ofeight kinds of hand motion modes can still achieve more than85%and the recognition rate of ifve kinds of hand motion modes more than96%. |
PDF Full Text Request