Estimation Of Muscle Force For Upper Limb Based On Wavelet Analysis Of SEMG Signals

The surface electromyography(sEMG)signal is a motor unit action potential trapping(MUAPT)superimposed on the skin surface,which is generated by the motion unit recruited at the time of muscle contraction,and it is a non-stationary signal.At present,the methods of extracting and classifying of the sEMG signal are not mature,especially for the electromyography control of the multi-functional intelligent upper limbs.The disabled is more likely to take the initiative to adjust the size of intelligent upper limb force,while autonomously controls the intelligent upper limb movement.In this situation,this thesis,studies the relationship between the characteristics of sEMG signal and the muscle strength of upper limb.At present,there are still many problems for the analysis of sEMG signals,for examples,sEMG signals in the collection process will be affected by the surrounding environment and the noise of the acquisition device,and the sEMG signal must be pretreated before analyzing.Due to the weakness of sEMG signals,fusion and classification of multiple features should be taken into account when extracting features;The general feature classifier has a low recognition rate for complex motion patterns,which affects the classification results.Therefore,this thesis finds a more appropriate processing algorithm to optimize the control signal source.First of all,a reasonable experimental program is developed,and then the collected sEMG was detected,denoised and the active segment was extracted,and the better classifier was selected to classify and recognize the muscle force pattern.First,a Gaussian mixture model is proposed to apply into sEMG signals in this thesis.In order to detect the start,stop,and interval of muscle activity,and the original sEMG signals are preprocessed to obtain the effective signals by the method of noise reduction in wavelet transform.Effective signal was obtained by the method of noise reduction in wavelet transform to preprocess the original sEMG signals.Then,the characteristics of sEMG signals were extracted from the time domain,frequency domain and time-frequency domain,and the relationship between the eigenvalues and the different muscle forces were analyzed quantitatively.It is difficult to obtain the relationship between the characteristic values and the muscle forces from the point of view of the time domain and the frequency domain.However,the time-frequency characteristics based on the wavelet analysis has a certain relationship with the strength of the muscle force,especially singularity feature vectors of the high and low frequency of wavelet coefficients have difference under different muscle strength levels.Finally,in pattern recognition,three classifiers of BP neural network,LDA linear discriminant and support vector machine are selected.Then the extracted feature vectors are input into each classifier,and five different muscle forces are used as the action identified method to classify and recognize.Through comparison and analysis of the classification results,it is found that the support vector machine is more efficient.