Research On Recognition Of Combination Motions Based On Timing Sequence Signal Of EMG

Biological EMG signals contain information such as muscle activity intensity and activity time,which can be used as input signals for human-computer interaction.The motor intention generated by active rehabilitation training of patients with hand dysfunction detected was output as control instruction to improve the training efficiency of patients.At present,most of the hand motion studies based on EMG signals are to identify isolated movements,while the purpose of rehabilitation training is to enable patients to carry out daily activities.Actual daily movements are completed by the combination of multiple isolated movements,but because EMG signals are non-stationary signals with high noise,it is difficult to identify the movement conversion process.In this paper,the EMG signals of hand movements were collected,the signal characteristics of EMG signals were studied,and the deep learning network was built for hand movement recognition.The combined movement recognition research based on EMG timing series signals was realized,with the expectation of improving the rehabilitation training effect of patients with hand dysfunction.The main research contents of this paper are as follows:(1)The EMG signal generated by the eight muscles of the arm during the hand movement is collected.Aiming at the eight-channel data,the method of wavelet transform was adopted to extract the time-frequency and time-frequency information of the eight-channel electrical signal data,and the time-frequency images were obtained.Finally,a single hand motion data set was established.(2)The recognition network is constructed to extract the global spatio-temporal features from the EMG time-frequency images of a single action of the hand.The network is a neural network structure combined with 3D Convolutional Neural Network(3DCNN)and Convolutional Short and Long Time Memory Network(ConvLSTM).Then the recognition network was used to identify and classify the single original motion of hand.(3)Design the combination sequence of the original hand movements,and collect the EMG signals of the daily actions after the combination.The resting state and motion state of the movement were obtained by analyzing the data with the method of detecting the movement segment.Then,the sliding window method is used to segment the eight-channel EMG signals in the combined motion and transform them into the time-frequency image stream of the combined motion.The established 3DCNN-CONVLSTM network model was used to identify ten original movements of the designed combined hand movements.Considering the signal overlap of the original action in the collection of the combined action,this paper adopts the majority voting algorithm to optimize the data set.(4)Baseline drift processing method is adopted to solve the problem of signal drift when hand EMG signal is collected in the actual interactive experiment.At the same time,the wavelet transform algorithm is used to filter the real-time data and optimize the conversion effect of the time-frequency image.Unity5.3.1F1 software was used to establish the experimental simulation environment of hand virtual reality interactive experiment,and the car in virtual reality was controlled by hand movement in the designed road environment.In this paper,the 3DCNN-CONVLSTM network model is used to study the recognition of combined actions based on EMG timing sequence signals.The experiment proves the effectiveness of the proposed method.