Improved EMD And IMF Energy Entropy Based Single Trial EEG Signal Classification

Brain-computer interface(BCI)technology is a research hotspot in the filed of brain science.BCI establishes a communicating channel between brainwave and external world.Through the brainwave controlled equipment,the subject can interact with the environment successfully.The emergence and development of the BCI technology extended the vision of researchers and encouraged their further study of brain functions and related working mechanisms.Previous studies have proved that motor imagery based BCI is a highly practicable system,yet its performance still needs further improvement,which might be done by optimizing the data processing,pattern recognition methods and the scheme design.For the purpose of improving the classification accuracy of single trial Electroencephalogram(EEG)signal during MI process,this study proposed a classification method which combined Intrinsic Modal Function(IMF)energy entropy and improved Empirical Mode Decomposition(EMD)scheme.Singular value decomposition(SVD),Gaussian mixture model(GMM),EMD and IMF energy entropy were employed for the newly designed scheme.The processing results were compared with that of the traditional EMD and Ensemble Empirical Mode Decomposition(EEMD).During the data acquisition,the subject was asked to imagine a target direction and watch the screen on which a sequence of arrows indicating directions were presented randomly.Once the the target direction shows up,the subject press down the corresponding button.The brainwave response to the target stimulation will give us a P300 signal due to the so called Oddball Paradigm,which would have been marked by the button press.The EEG signals were recorded from 32 subjects using the "Neuroscan EEG Acquisition System".After removing noise and artifacts in EEGLAB,Singular value decomposition was applied,and the singular values were clustered by Gaussian mixture model.The insignificant characteristics indicated by the small SVD values were then removed,and the signal was reconstructed,feeding to EMD algorithm.Not surprisingly,each IMF component was mainly consisted of a signal of certain frequency,those IMFs mapping to δ、θ、αand β frequencies were selected as the major features of the EEG signal.The SVM classifier with RBF,linear,and polynomial kernel functions kicked in.A total of five SVM classifiers were constructed for the five directions(up,down,left,right,stop).The testing signal(unknown signal)passed through each of the five classifiers,and their votes will determine the category of the incoming signal.The results showed that the optimized scheme greatly improved the classification accuracies,suggesting the probability of designing more efficient EEG control system based on the proposed scheme.Furthermore,this study also noticed that the energy and energy entropy of β rhythm highly negatively correlated with the attention level of the subject,which was directly associated with the EEG classification accuracy,and which should be considered in future EEG classification system design.