Research On Time-Spatial-Frequency Domain Finger Motion Recognition Based On SEMG

Hand is one of the most flexible organs of human body movement,and hand movement research has always been a research content worthy of attention in the field of rehabilitation medicine.The current intelligent artificial hand only simulates simple actions such as closing and opening of the limb,and it is impossible to simulate the fine motion.However,the high density surface EMG(HD-sEMG)signal acquisition system is adopted,which can overcome the shortcomings of traditional surface electrodes that cannot identify muscle group sEMG signals.More channels can collect the muscle electrical signals covering the defined skin area,and capture the time domain and spatial domain distribution information of the myoelectric signal of the whole muscle activity area,which is beneficial to the analysis of the detailed action EMG signal.Since the human hand movement,especially the analysis of the myoelectric signal of the finger is difficult.An HD-sEMG signal analysis method of the finger refinement motion is proposed in this thesis.The HD-sEMG signals of flexor digitorum superficialis flexors are analyzed for four different bending angles(15°,45°,70° and 90°)of the experimenter’s fingers.In this thesis,firstly,the 16-channel HD-sEMG electrode is placed on the flexor digitorum superficialis for HD-sEMG signal acquisition.In terms of data preprocessing,principle components analysis(PCA),fast independent components analysis(Fast ICA)and multiclass common spatial patterns(multiclass CSP)mode are applied to filter the collected HD-sEMG,which can reduce redundant information and data dimension,and then obtain a separation matrix to reconstruct the original information.In order to reduce the number of channel and achieve the purpose of filtering the channels with strong muscle electrical signals from the original channels a channel selection method for maximizing the mutual information is proposed based on the multiclass CSP,the signal channels are arranged in descending order according to mutual information maximization.Secondly,in the feature extraction,the time-domain,frequency-domain,time-frequency domain features and time-frequency domain features decomposed by wavelet packet transform of the pre-processed HD-sEMG signals are extracted.In addition,the time-frequency domain features decomposed by wavelet packet transform(WPT)are extracted,which combine with the spatial features of the sEMG signal extracted by the channel selection algorithm based on mutual information maximization.By comparing and analyzing the above feature sets,it is found that the feature set combined with the time-space-frequency domain features have more distinct distinguishing features on the four action modes of the finger.Finally,linear discriminant analysis(LDA),artificial neural network(ANN)and support vector machine(SVM)are trained by feature sets of time-frequency domain features and time-space-frequency domain features respectively.When five original signal channels are selected,the pattern recognition results show that SVM based on the combination time-spac-frequency domain features can achieve the expected recognition accuracy of 86.7%.