Research On Decoding Of Human Hand Movement Intention Based On SEMG Signal

The decode of human hand movement intention is to decode the desired motion expression by using the information of the motor nerve potential of the surface cortex.The rationale behind this technique is that when a person have a certain behavioral intention,the cerebral cortex produces corresponding motor neuron information,which can be obtained by decoding the surface electromyogram(sEMG)signal.A variety of motion information corresponding to human hand motion is decoded from the sEMG signal to ensure that it has sufficient classification accuracy so as to establish the mapping relationship between the sEMG signal and the motion information.These theories and research can be applied to the rehabilitation of hemiplegic patients.The patient can receive effective rehabilitation training on the limbs by wearing wearable gloves controlled by sEMG signal.Based on the physiological basis of sEMG,this paper analyzes its characteristics and mathematical model,as well as the choice of muscle and electrode during the test.The circuit of the hardware acquisition board is analyzed in detail and A / D conversion is performed by ARM.A software interface based on Matlab that can acquire EMG,waveform display and data storage through serial port in real time is designed.Secondly,pretreatment of the collected sEMG signal.This paper selects the wavelet transform to denoise the sEMG signal.Described in detail the sEMG signal denoising criteria,process and threshold selection,and using sym8 wavelet basis,the original signal decomposition of 4 layers.Finally,it is verified by experiments that the minimax threshold rule is better for noise reduction.Meanwhile,existing human motion intent decoding methods cannot extract in-depth source information implicit in the observed signal and are only applicable if the signal to noise ratio is high and the waveform overlays are low.Considering these shortcomings,this paper presents a decoding method based on motor unit action potential train(MUAPT).The proposed method uses one of blind source separation algorithm fast independent component analysis(FastICA)based on negative entropy to extract the implicit MUAPT information,and extracts the features by locality preserving projection(LPP)algorithm.And,it is classified by hidden markov model(HMM)and finally decoded six kinds of hand intent actions.In addition,the method can effectively improve the recognition rate of the human hand movement intention by the experimental results.Finally,by designing the control system part,the rehabilitation training of the wearable rehabilitation robotic hand is successfully achieved.This part adopts pneumatic control mode,and uses ARM microcontroller to output pulse width modulation(PWM)signal with adjustable duty ratio.It controls the opening and closing time of high speed on-off valve,and makes the pneumatic muscle stretch and move,so as to complete the training action of connecting rod to drive the robot's hand joint to grasp.Combined with the "meditation training method",the "human-computer interaction" can be realized to speed up the reconstruction of the patients' nerves and improve the efficiency of rehabilitation training.