The Recognition Of SEMG Hand Actions Based On SVM

The surface electromyography signal is a kind of bio-electricity signal thatgenerates from the skeletal muscle and collected on the surface of human’sbodies by the EMG acquisition instrument. sEMG has the feature of convenience,accuracy and non-invasive test, so it has been used in the field of rehabilitationmedicine, sports medicine, and the intelligent robot. However, the control ofbionic rubber hand recognition is still not mature enough, there is some distancebetween theoretical research and practical applications.In this thesis, we can collect sEMG of hand actions based on optimal SVMin order to control the bionic artificial limbs more rapidly and sensitively. Themain points contain: sEMG collection, sEMG pre-processing, and sEMG handactions recognition.We first analyzed the research status of bionic artificial limbs and sEMGhand actions recognition in detail and studied its physiological propertyspecificly. Thus we ensured the four major muscles groups on the arm as theoptimal muscle positions during the signal collection. And we completed sEMGextraction of common hand action patterns based on the patch electrodes andelectromyographic signal acquisition instrument after analyzing andprogramming to the hand actions systematically.In the preprocessing, we denoised the sEMG for the clearer analysiswaveform through the wavelet functions. Then we extracted the denoised-signalfeatures based on the time domain analysis, frequency domain analysis, and time-frequency domain analysis, after comparing the extracted datas we chosed time-frequency domain features based on the wavelet packet function to be itseigenvector benchmark and selected the variance and energy of the waveletpacket coefficient as its feature vectors.In pattern classification, we selected two major recognition classifiers: SVM classifier and LSSVM classifier, we input the feature vectors to the classifiers,and recognized the sEMG hand actions after the analytical processing. Then weanalyzed the correct recognition rate, and the training time of two kinds ofclassifiers under the parameter optimization after the discussion and experimental,and then obtained that the pattern recognition under LSSVM based on the PSOhad a higher recognition rate and a shorter operation time.