| The electromyographic signals of human body contain abundant motion information,and the electromyographic signals are generated before the actual motion of human body,which can provide prediction of human body motion,this feature can be well applied in the sensing module of flexible booster suit system.At present,the human motion intent sensing technology for flexible booster suits mainly uses traditional physical sensors to collect human motion information,which has certain data lag and cannot achieve more complex motion intent understanding.In order to solve the above problems,the surface electromyography(SEMG)signal is used to realize the perception and recognition of human upper limb action intention.The specific research contents are as follows:First,the motion of human upper limb shoulder,elbow and wrist joints were analyzed,the mechanism of EMG signal generation was explored,and the experimental scheme of EMG signal acquisition was determined.Aiming at the problem that the EMG signal itself is weak and the upper limb muscles are close to the heart,it’s easily interfered by the ECG signals and other noise signals when it is collected,this thesis proposes a combination of IIR filtering based on empirical modal decomposition(EMD)and wavelet transform for noise removal,and extracts the time domain and frequency domain features of the EMG signal.The results show that the signal-to-noise ratio,root mean square,and autocorrelation coefficient of the signal after filtering and noise reduction using the method in this paper are better,and the denoised signal is smoother and has better noise suppression effect.Secondly,different classification models of action intention are designed to classify and recognize 9 types of action intention patterns of human upper limbs.The recognition accuracy of naive Bayesian cognitive science was only 75.67%,that of Multi-Layer Perceptron(MLP)and Support Vector Machine(SVM)was about 93%.On this basis,the Grey wolf optimization algorithm(GWO)is used to optimize the SVM,and the recognition accuracy of the optimized GWO-SVM model reaches 99.78%.Two methods,principal component analysis(PCA)and t-distribution random neighborhood embedding(TSNE),were used to analyze the effects of different dimensions on classification accuracy,the results show that the optimized GWO-SVM model has the highest recognition accuracy in different feature dimensions.Finally,the software of action intention recognition based on surface EMG signal is designed and implemented,and the practicality and feasibility of the software are experimentally verified.In view of the low efficiency and single function of the signal processing software accompanying some current commercial sensors,thesis uses the APP Designer module of MATLAB to design a software integrating EMG signal preprocessing,feature extraction and classification recognition,which simplifies the process of EMG signal processing and action intention recognition.The software package after transplantation was used to predict the EMG data sets of 9 kinds of human upper limbs.The accuracy of GWO-SVM model was 96.87%,the feasibility of the GWO-SVM model based on the surface EMG signal is verified in the actual action intention prediction. |
