Research On Motor Imagery Brain-Computer Interface Based On OVR-CSP And Cross-Correlation

Brain-computer interface(BCI) isa communication control system which allows direct translation of brain states into actions, bypassing the usual muscular pathways. After decades of development, BCI has not only been applied to help with the communication of the disabled, but also to the entertainment, environmental control and medical rehabilitation and other fields. In short, improving the quality of life of people with disabilities has always been an ultimate purpose of the study of BCI. Motor imagery(MI) is considered to be the most suitable electroencephalography(EEG) for the EEG based brain-computer interface, because MI is a kind of spontaneous EEG which needs no more extra device.In this paper, based on the One-Versus-Rest Common Pattern Spatial(OVR-CSP) which is well known for the feature extraction of four-class motor imgery, a new improved OVR-CSP algorithm is proposed. The features based on OVR-CSP which is a construction of signal variance reflects the energy changes in the corresponding regions of the brain. However, OVR-CSP based features does not contain information about the cooperation and information exchange between different regions of the brain. And the information is conducive to correct classification. By using the method of simultaneous analysis, this kind of information can be obtained quickly. And the cross-correlation function is an effective method to analyze the synchronization. Therefore, feature based on OVR-CSP and feature which is extracted by cross-correlation function are fused to obtain the feature for classification in feature extraction stage of this paper. Then, the feature for classification contains both information about energy changes and the synchronization information between channels of somatosensory area, which are complementary. Experimental results on the public data set of the second international BCI contest show that the average Kappa value of the proposed feature extraction algorithm is highest, reaching 0.59, with the smallest variance 0.031. This has fully demonstrated that the feature extraction algorithm proposed in this paper not only has good classification performance, but also has a strong adaptability to individual differences. On the basis of the above, the performance of the windowing in proposed algorithm is also discussed.In consideration of the difficult and cumbersome process and the greatly limited application situation of existing BCI, we use the portable EEG device Emotiv EPOC+ to carry out the acquisition of motor imagery EEG signal. Then, with the help of the feature extraction algorithm proposed in this paper, we have established a MI BCI based on Emotiv EPOC+. This MI BCI is a picture browser based on EEG, and the picture is selected through the four types of motor imagery, the imagination of left hand, right hand, foot and tongue. After off-line test, the proposed BCI turned out to be a MI BCI with good performance, which has laid the necessary foundation for the future development of brain computer interface.