Research On Gesture Recognition Based On Surface Electromyography Signal

In the field of human-computer interaction,gesture recognition based on surface electromyography(s EMG)is a key component of hand rehabilitation training systems and a research hotspot in the field of rehabilitation robotics.However,most existing signal processing methods still have limitations in the filtering and feature selection of s EMG signals,mainly manifested in the following aspects: the filtering methods can only eliminate one or two specific types of noise,and it is difficult to achieve an effective balance between the selected feature dimensions and the final prediction accuracy in feature selection.To address these issues,this study selected five representative participants to collect surface electromyographic signals for six grasping gestures,and conducted a comprehensive analysis using a self-developed novel filtering algorithm and feature selection algorithm,ultimately improving the accuracy and stability of the gesture recognition model.The main research contents are as follows:(1)In response to the problem that surface electromyographic(s EMG)signals are easily affected by various noise interferences during the collection process,a denoising scheme based on synchronized wavelet transform(SWT)algorithm was designed.The results showed that after denoising with this method,the power line interference(PLI),baseline drift(BW),and Gaussian white noise(WGN)in the s EMG signals are effectively suppressed,providing high-quality data support for subsequent algorithms.(2)To effectively balance the contradiction between feature dimension and final prediction accuracy,a novel fusion-based feature selection algorithm called Relief FBPSO-KNN(RBK)was proposed.This algorithm organically combines Relief F,BPSO,and KNN algorithms,utilizing the advantages of each algorithm to compensate for their shortcomings,and thereby improving the feature reduction ability and prediction accuracy of the algorithm.In addition,this thesis also proposes an empirical formula based on adaptive class Gaussian distribution,which improves the problem of the RBK algorithm being prone to premature convergence and falling into local minima.(3)To verify the correctness and effectiveness of the proposed algorithm,signalto-noise ratio,root mean square error,and R-squared value were used to evaluate the filtering algorithm,classification accuracy and number of features were used to evaluate the feature selection algorithm,while one-way analysis of variance is used to verify the advantages of the proposed algorithm.The results show that the filtering and feature selection algorithms proposed in this thesis are significantly better than other similar algorithms in most samples(the value of one-way ANOVA p<0.05).Moreover,based on the proposed algorithm,this thesis designs a gesture recognition scheme based on s EMG signals,which is applied to a self-developed hand rehabilitation robot platform in the laboratory.The results show that the proposed scheme has an average classification accuracy of up to 95% for six hand gestures that can be used for rehabilitation,providing a new feasible solution for the practical application of hand function rehabilitation.