Recognition Study On Left & Right Arm MI EEG

Brain Computer Interface (BCI) is a new technologytoconnect human with computer, which could be used in many fields including helping the disabled to live independently. There are three essential parts in BCI which includingsignal acquisition and signal processing and controller.Signal processing is an important part of the three ones.This article analyzes EEG produced by motor imagination of left & right arm.The data is downloaded on the website of BCI competition IV, which is contributed by the BCI research group in Graz University of Technology.There is an introduction about BCI in the beginning.Present development situationand main research methods of BCI are included similarly.During the part of signal preprocessing, this article has chosen appropriate channels and appropriatefrequency bandto further studyaccording to spectrogramsand brain electrical activity mappings of alpha energycoming from different subjects and the current research achievements. The phenomenon of event related desynchronization&synchronization are presented.During the part of feature extraction, EEG signal of motor imagination is divided into several sections in order to calculate energy spectrum and forth order cumulantrespectively. Hurstexponentand phase locking value are imported to characterizenonlinear and non-stationary EEG. This article investigates advantages and disadvantages of each feature. Accuracies of pattern recognition based on single kind of features and jointfeatures and features which have been reduced in dimensions by principle component analysis are comparedsubsequently. This article imports random forest classification in the part of pattern recognition while linear discriminant analysis and support vector machine are imported in the same way. Influences on classification accuracy which is caused by different parameters of different classifiers are investigated as well in the part.Considering the low classification accuracy,Common spatial patternwhich has been combined with wave packet transform and empirical mode decompositionis utilized to extract the features. This article has achieved accuracy of 96.92% in single subject and 75.52% around all the subjects under the situation that the ERS/ERD phenomenon is not obvious.There are several innovations in this article.For the first,curve of event related potential is proposed to describe the phenomenon of ERS/ERD which is simple in calculation.For the second, EEG signal is divided into several sections to calculate energy spectrum and forth order cumulant. Hurst exponentwhich is usually used to describe longrange correlationis imported as a new feature. Random forest classification is employed in the pattern recognition part.For the last, common spatial pattern combined with wave packet transform and empirical mode decomposition is used to extract features. The accuracy has achieved 96.92% in single subject and 75.52% around all the subjects under the situation that the ERS/ERD phenomenon is not obvious.EEG analysis is a research fieldwhich is full of complexities and challenges. This article has conducted a relative-depth exploration and made some contributions to the study of motor imagination EEG.