Application Of Eeg Source Imaging In Classification Of Motor Imagery Tasks

The rapid development of electroencephalogram(EEG)technology has opened up a space for the wide application of motor imagery(MI)in brain-computer interface(BCI).Unlike the induced EEG signal,motor imagination EEG is an endogenous spontaneous signal that requires only the imagination of the subjects with ideas without external stimulation.Because of the simple,flexible and non-invasive features of this technique,there are many application scenarios,which can be used not only for rehabilitation physiotherapy but also for rehabilitation physiotherapy for leisure and entertainment and 50 on.In this paper,an innovative method has been proposed to solve the problem of low classification accuracy and poor generalization performance of multi-task MI by combining scout and the convolutional neural networks(CNNs)based on EEG Source Imaging(ESI).The classification problem was optimized.Firstly,the EEGs need to be preprocessed to improve the signal-to-noise ratio.The notch filter is used to remove the power frequency of the actual environment during data acquisition.Independent component analysis(ICA)is used to remove eye artifacts.The bandpass filter is used to pass the band 8～3 0Hz,only paying attention to the frequency band related to MI.Then,considering that the data measured by the EEG is distorted and not intuitive’ enough,ESI technology is used to convert the data of the sensor domain to that of the source domain,which is to project raw data into the cerebral cortex.The head model is created by boundary element method(BEM),and the inverse problem is solved by weighted minimum norm estimation(WMNE).In the absence of magnetic resonance imaging(MRI)data,the Colin 27 head with high resolution and quality was used.The motor cortex is closely related to Ml,and the region of most interest studied in this paper is selected from the motor cortex.According to the Penfield’s homunculus,different areas of the motor cortex correspond to different limb movements.10 scouts are created in the motor cortex,each of which contains 40 sources,ensuring that the scouts have the same size.The joint timefrequency analysis(JTFA)of the scouts’ time series is performed by Morlet wavelet.Convolutional neural networks(CNNs)are widely used in image processing and perform well.Since the time-frequency diagrams are pseudo-color images,CNNs are proposed to use for classification.A time-frequency diagram is used as the input to the CNNs,and the output is the probability of each of the 4 classes.The classiflcation results show that we report an increase of up to 14.4%for overall classification compared with the competitive results of the state-of-the-art MI classification methods.At the same time,based on the ESI,the combination of the scout and CNNs is applied to the classification of EEG signals,which also provides a new idea for dealing with the classification of high-dimensional EEG signals.