Research On EEG Signals Artifacts Elimination And Feature Extraction Of Brain Controlled Robot

The brain computer interface based on motor imagery electroencephalogram(EEG)can provide a communication approach between human body and outside world without relying on peripheral nerve or musculature.The brain-controlled robot combines robot controlling technique with brain computer interface so that human brain can accomplish certain movement by EEG signals.Though EEG signals are weak,random and sensitive which can be easily affected by artifacts like eye electric signal,cardiac noise etc.Thus,one of the greatest challenge for brain-controlled robot to react quickly and correctly is how to extract EEG features of motor imagery.Since EEG signals varies from one person to another which leads to low accuracy of motor imagery EEG signals classification.Based on problems raised above,it is necessary to study artifacts elimination and feature extraction of EEG signals.This paper studies group-information guided independent component analysis of motor imagery EEG signals artifact elimination,as well as brain network based multi-domain fusion of multiscale feature extraction.In addition,eigenvector is optimized to improve the accuracy of multiclass motor imagery classification.A brain-controlled robot experimental platform is designed and implemented based on motor imagery EEG signals.The main work is as follows:(1)Considering the interference from other artifacts in EEG signals,a group-information guided independent component analysis is proposed.It aims to allow multiple decomposition of the independent component has the correspondence among different subjects,reduce the sensitivity of the reference value selection and improve the robustness of the algorithm.The results show that compared to classic independent component analysis with references,the proposed method can separate artifact component from EEG signals efficiently so that the classification accuracy of feature extraction is improved significantly.(2)Considering the differences in EEG signals among subjects,which normally lead to low precision of motor imagery classification,a brain network based multi-domain fusion of multiscale feature extraction of motor imagery EEG signals is proposed.Complex network analysis method is applied to measure the characteristics of brain network based on brain functional connectivity.The topology diagram of brain network function is built.Based on which different brain network measure indicators are extracted.The experimental results show that compared with the common space mode algorithm and the common spatial mode combined with local characteristic-scale decomposition algorithm,this algorithm can perform the classification of motor imagery EEG signals more efficiently.(3)A brain-controlled robot experiment platform is designed and realized.Based on methods raised above,the interactive experiment platform which applies brain-computer interface and NAO robot is constructed.The acquisition and processing of real-time EEG signals module as well as a human-computer interface is established.The motor imagery EEG signals are recognized as 4 motion modes.The proposed algorithms are verified through interaction of EEG signals and online brain-computer interface system.