Evaluation Of Lower Limbs Rehabilitation Based On Time Domain Feature Analysis Of SEMG Signals

Dystonia is a kind of sports disorder syndrome with high incidence at present.This kind of disease not only brings inconvenience to the patients’ life,such as unable to take care of themselves,but also seriously affects the patients’ physical and mental health.Traditional clinical rehabilitation has many disadvantages because of the limitation of both doctors and patients.The combination of sEMG signal and clinical treatment can qualitatively evaluate the diagnosis of subjects through data analysis,thus greatly improving the training efficiency.However,as the basis of data analysis,signal acquisition may lead to loss of some channel signals or errors due to cumbersome acquisition process and expensive acquisition equipment.Therefore,this paper proposes to apply the neural network models to the sEMG signals prediction of the missing channel,and use the predicted sEMG signal to extract characteristic information for the lower limb rehabilitation evaluation and analysis.The main work of the paper is as follows:First,the overall scheme of the experiment was designed according to the generation mechanism and characteristics of sEMG signals,including determination of electrode type,muscle tissue and electrode placement,healthy subjects,patient subjects and typical movements of lower limbs,and the acquisition of sEMG signals of each muscle channel was completed.Secondly,the noise sources of external interference in the acquisition of sEMG signals were analyzed,and the signal was filtered by butterworth low-pass filter.Based on the Keras deep learning framework,this paper built a deep neural network(DNN)model,support vector regression(SVR)model and a stack autoencoder(SAE)model to predict sEMG signals of the rectus femoris(RF)muscle of healthy subjects.Finally,the analysis and processing of sEMG signals is introduced,including time domain feature extraction and significant difference analysis.By extracting the feature information of each typical movement of prediction data of healthy subjects and the collected by the patient subjects,and using the t-test of significance level = 0.05 to analyze the significant difference between the variation coefficient of each feature of the two groups of subjects,the diagnostic evaluation of dystonia was realized.In this paper,the neural network models are used to predict the missing channel sEMG signals and obtain good experimental results.Therefore,the proposed neural network models can be effectively used for EMG capture of finite channels and calculation of missing channel signals,which not only saves time and energy,but also saves the cost of capture equipment.At the same time,the method of time domain feature extraction and significant difference analysis can also effectively distinguish the diagnosis between the two groups of subjects,thus laying a foundation for future research.