Research On The Analysis Of Exercise-induced EEG Signals

Rehabilitation robot can help the paralytics and disabled elderly in their exercise of rehabilitating the motor functions.Directing at a kind of lower limb rehabilitation robot which is based on the principle of stretch reflex,this thesis studies the processing and recognition of electromyographic(EMG)and electroencephalograms(EEG)signals respectively in the evoked and non-evoked reflex motion exercise modes,and analyzes the influence of rehabilitation robot's motion evoking on neural activity.This thesis mainly focuses on the following aspects:(1)A multi-modal biological information acquisition platform is established;the electromyographic(EMG)and electroencephalograms(EEG)signals in both evoked and non-evoked motion modes are acquired by taking healthy postgraduates as the subjects;the EEG signals in the effective frequency range of 8-30 Hz is obtained through the band-pass filtering processing on EEG signals.(2)In order to solve the mode mixing phenomenon during traditional signal decomposition and the over or insufficient decomposition in traditional signal decomposition algorithm,the Variational Mode Decomposition(VMD)algorithm is adopted in this thesis to decompose the signal and the intrinsic modal component of the signal at the best K value is obtained.(3)The Hilbert Transform is applied to calculate the signal's marginal spectrum and instantaneous energy spectrum;based on taking the amplitude mean,amplitude variance and instantaneous energy mean as the EMG features of the two exercise modes,the stretch reflex in evoked exercise is judged via the EMG features.The sample entropy under different scale factors,and the Multiscale Entropy(MSE)with the largest entropy value as well as the most obvious discrimination are selected as the non-linear dynamic features of the EEG signal;the amplitude mean,amplitude variance,instantaneous energy mean and MSE are used as EEG signal features.(4)Considering that it is unable to determine the initialization parameters of traditional support vector machine,the Grey Wolf Optimizer(GWO)support vector machine is used in this thesis,so that optimized input penalty factor and kernel function are obtained.The optimized support vector machine is more efficient in indentifying the motor imaging EEG signals in international BCI competition data.(5)In the analysis of exercise-evoked EEG signals,experimental study on the EEG signals of 5 subjects under evoked and non-evoked exercise trainings is conducted.The traditional support vector machine is used to judge the recognition of stretch reflex by EEG signals in different feature sets,and the feature set and classifiers with the best recognition rate is obtained through the statistical analysis on the recognition results.The experimental results show that the proposed method can effectively analyze exercise-evoked EEG signals.