The Research On Fine Motion Pattern Recognition Of Upper Limb Based On Surface Electromyography

The surface electronomygraphy signal(sEMG)carries a great deal of information about patterns of movement related to the human body.With the development of biomedical technology and pattern recognition technology,sEMG has been widely used in medical assessment,artificial limb control,human-computer interaction and so on.This paper selects eight kinds of hand movements which are based on the survey of ten hand injuries from Guangdong Work Injury Rehabilitation Center.And then we select the superficial flexor,thumb long flexor,referring to the total extensor,ulnar flexor to place EMG sensors according to the selected action and arm muscle structure.In this way,we can collect the sEMG we need in the experiment.As for sEMG processing and analysis,the main work of this paper focused on signal preprocessing,feature extraction,dimension reduction and classification.The specific steps are shown below:(1)Signal preprocessing.The sEMG preprocessing mainly includes two aspects: noise elimination and activity segment extraction.On electromyographic signal de-noising,Sym8 is chosen as the basis function of wavelet threshold de-noising.Compared with the traditional butterworth filter,this method can obtain ideal de-noising result.In the activity segments extraction,the use of variable steps moving average method can work good not only achieve the function of extracting the active segments correctly,but also speed up the calculation.(2)Feature extraction.In this paper the representative features in the time domain,frequency domain and time-frequency domain are selected to study,and then use the DB index to evaluate the features.Taking the result of DB index as the criterion of selecting features,we can meet the classification requirements through experiments.(3)Dimension reduction and classification methods.Different combinations of dimensionality reduction methods and classifiers can produce different classification results.Two kinds of dimensionality reduction methods: principal component analysis(PCA)and local linear embedding(LLE),two classification methods: support vector machine(SVM)and linear discriminant analysis(LDA)are selected.They can combine four classification methods: PCA-LDA,PCA-SVM,LLE-SVM and LLE-LDA.In this paper,the above four classification methods are studied,and the different components of support vector machines in solving the multi-classification problems are compared and analyzed.