Reaserch Of Robotic Control Based On Motor Imagery EEG

Brain-Computer Interface (BCI) is a novel interactive way which has attractedwidespread attention and research by scholars at home and abroad. BCI establishes acommunication system between the human brain and peripheral for transportinginformation rather than depend peripheral nerves and muscles. Thus, auxiliary devicescan be controlled by EEG to provide the disabled person with service, regulateemotions of people suffering from anxiety disorders, assist the rehabilitation trainingof hemiplegia patients, and provide new forms of entertainment for people.In this paper, the problems of Motor imagery EEG preprocessing, featureextraction and pattern classification are analyzed and a method that the EEG signal isused to control the robot motion is designed. In this paper, research works werearranged as follows:(1) In order to eliminate the noises in EEG, an improved threshold estimatingmethod is adopted to denoise. In improved threshold estimating algorithm, thewavelet coefficients are estimated by processing the power of the wavelet coefficientsand the power of the threshold. It can not only overcome the shortcomings oftraditional wavelet thresholding methods, but also retain advantages of them.(2)When imagining unilateral limb movements, ERD/ERS phenomena of theMotor imagery EEG in specific frequency band are more significant, therefore richinformation of Motor imagery tasks can be obtained. In addition, because ofcharacteristic of non-linear in EEG, a feature extraction method of combiningDual-Tree Complex Wavelet Transform and sample entropy is presented. The EEG isdecomposed into multilayer by DTCWT and corresponding band of ERD/ERSphenomenon in the Motor imagery EEG signal was extracted from decomposed signal.Then the rhythm signal is extracted feature extraction using the sample entropy.(3)In the applications of hidden Markov model (HMM) in motor imageryelectroencephalogram (EEG) classification, the independence assumption of HMM isinconsistent with the inherent correlation of EEG signals. In order to resolve theproblem, an EEG classification method based on Choquet fuzzy integral HMM(CI-HMM) is proposed. The independence assumption of HMM is relaxed bysubstituting the monotonicity of fuzzy integrals for the additivity of probability measures. This method can effectively improve the performance of the HMM methodused in Motor imagery EEG classification.(4)An experimental program of BCI system is designed for the control of robotwith motor imagery EEG signal. Four motion imagery tasks, i.e. kicking ball withright foot, picking ball with left hand, hopping with left foot and simultaneouslythrowing the ball with left hand, hopping forward with right foot and then picking ballwith right hand, are used as control command for four basic movement of roboticincluding forward, brake, turn left, turn right.(5)The2008BCI Competition IV Datasets1data and experimental data ofmulti-class Motor imagery EEG are used to make experiments. The experimentalresults show that feature extraction method of DTCWT sample entropy caneffectively improve discrimination of the different types of motor imagery EEG. Forclassification of multiple types of complex Motion imagery EEG, the recognition rateof CI-HMM classifier is higher than SVM classifier.