Motion Intention Recognition And Influence Analysis Of Force Change Based On EMG Information

EMG pattern recognition is an advanced intelligent signal processing technology and has been regarded as an important method for reliable user intention classification.Relevant studies have reported high classification accuracy in the laboratory,but they are not satisfactory in clinical application.One important reason is that the EMG-PR method is still not robust enough to deal with many problems,such as electrode displacement,muscle fatigue,and change of force.Among them,the change of force is the key problem that affects the performance of EMG-PR method.Therefore,it is very important to improve the robustness of EMG pattern recognition method against changes of force.On the other hand,commonly used commercial prosthetics have only active motion control,but no active force control,which greatly reduces the willingness of amputees to wear prosthetics.The research of this thesis is mainly divided into three parts: motion intention recognition,grip force estimation and synchronous decoding of motion intention and force.For motion intention recognition,the main research is the feature of resisting the influence of force change.For the estimation of force,the regression model of EMG and grip force was established.For the synchronous decoding of motion intention and force,the synchronous decoding of motion intention,continuous grip force and discrete force is mainly discussed.The main contents and innovative work of the thesis are as follows:(1)Firstly,for the motion intention recognition under the influence of force,logarithmic discrete Fourier transform(DFTL)and normalized logarithmic discrete Fourier transform(gn DFTL)are proposed,and the linear discriminant analysis(LDA)algorithm is used for classification.The results show that,compared with the commonly used time domain feature,frequency domain feature and time-dependent power spectrum(TDPSD)features,the accuracy and robustness of gn DFTL features reach 95.24% and 2.24 respectively in healthy people,which are obviously better than time domain,frequency domain and TDPSD features.The accuracy and robustness in amputees also reaches 88.99% and 1.52,in which the accuracy and robustness of hand close(HC)reach 97.43% and 2.28 respectively,the robustness is obviously improved compared with the TDPSD feature,which lays a good foundation for the decoding of amputee grip force.(2)Secondly,for the problem of force estimation,the DFTL feature is modified,the meanch-DFTL feature is proposed,and the representative time domain,frequency domain and AR coefficient features are selected for comparison,and artificial neural network(ANN)was used for estimation.The results show that the relative errors of meanch-DFTL features proposed in this thesis in healthy and amputee are 0.065 and 0.108 respectively,and the correlation coefficients are 0.953 and 0.910 respectively,which are better than the other three features.(3)Finally,for the synchronous output of motion intention and continuous grip force,a fusion output scheme is designed.For the synchronous decoding of motion intention and discrete force,a convolutional neural network(CNN)model is designed for the synchronous decoding of motion intention and discrete force.Compared with LDA algorithm,the accuracy of healthy and amputee is improved by 1.3% and 3.61% respectively.