Research On Hand Movement Recognition Based On Mixed Feature Separation Of SEMG Signal

Surface electromyography signal(s EMG)is a kind of bioelectric signal that reflects the activity state of human muscles and nerves.And because of its simple and convenient extraction method,it is widely used in the field of rehabilitation medicine.The hand is one of the most complex limb structures of the human body.The classification of hand complex movements s EMG is always a difficult point in the field of rehabilitation medicine.Aiming at the classification of complex hand movements,this thesis proposes a recognition method based on s EMG hybrid feature separation.First,according to the analysis of the mechanism of s EMG generation,it is concluded that the feature separation of s EMG of complex movements is feasible.By analyzing the daily activity habits of human hands,the six most common complex hand movements are summarized and further analysis is performed.According to the analysis of the upper limb muscle distribution involved in the six movements,the appropriate s EMG signal sampling location is determined.And four-channel synchronous acquisition of forearm s EMG experiment is performed to provide a data basis for follow-up research.Secondly,in terms of s EMG preprocessing,the global threshold noise reduction and layered threshold noise reduction in the wavelet transform are used to de-noise the original s EMG,and the de-noising effect is analyzed,improving the fidelity of the EMG signal.In terms of feature extraction,s EMG is extracted from three aspects: time domain,frequency domain,and time-frequency domain.A total of 7 features are extracted.By observing the clustering distribution of features,it is difficult to directly train a single complex action classification model.Aiming at this problem,a classification strategy is proposed to separate complex hand movements into a combination of finger movements and wrist movements.The feature distribution of the separated movements has obvious separability,which solves the problem of the confusion of the feature distribution of complex movements.Finally,complex actions are classified based on the action feature separation strategy.Taking the features of the separated actions as input,build a finger action classifier and a wrist action classifier.Three classification methods are selected: K-nearest neighbor classifier,support vector machine and convolutional neural network.Comparing the recognition results of three classifiers when combining time domain,frequency domain,and time-frequency domain features,it is concluded that the use of convolutional neural network for pattern recognition of time-frequency domain features has the highest recognition rate.A "combined" classifier is proposed.The classifier trained with separate features is used as the sub-classifier,and the classifier results of the two sub-classifiers are reconstructed as the final classification result.In order to compare the classification performance of the "combined" classifier,the six hand movements are directly classified.The results show that the recognition rate of the "combined" classifier is significantly improved.In addition,an intelligent artificial hand control experimental platform is built for experimental verification.