Motion image-electroencephalograph (MI-EEG) recognition method based on brain-derived domain space

Abstract

The invention discloses a motion image-electroencephalograph (MI-EEG) recognition method based on brain-derived domain space. The recognition method comprises the steps as follows: performing common average reference and band-pass filtering and other preprocessing on acquired MI-EEG; performing inverse transformation on the EEG by standardized low-resolution brain electromagnetic CT algorithm to obtain a brain-derived domain dipole moment amplitude time sequence; primarily selecting dipoles on the basis of dipole moment amplitude by a data driven method, performing time-frequency analysis on the dipoles by continuous wavelet transform to select dipoles and determine optimal time period; and extracting dipole wavelet coefficient powder sequence characteristics by a one-to-one common spatialpattern algorithm, and inputting the characteristics in a support vector machine for classification. The time domain, frequency domain and spatial domain information are fully utilized in optimization of dipoles, determination of optimal time period and characteristic extraction while the spatial resolution is improved, and the method has great significance in improving calculation efficiency andclassification accuracy.

Description

technical field [0001] The invention belongs to the technical field of recognition and processing of motor imagery electroencephalogram (MI-EEG) signals based on brain source space, and specifically relates to the inverse conversion of scalp electroencephalogram signals into Cerebral cortex, based on data-driven (Data-driven) and continuous wavelet transform (CWT) to optimize dipoles and determine the optimal time period, and use one-to-one common space pattern algorithm (OVO-CSP) and support vector machine (SVM ) to realize the feature extraction and classification of MI-EEG in the brain source space. Background technique [0002] Brain-computer interface BCI (brain-computer interface) is not dependent on the conventional brain information output pathways such as peripheral nerves and muscle tissue, but uses engineering technology to establish a connection between the brain and computers or other electromechanical devices to achieve "letting the mind A new way of external ...

AI Technical Summary

Problems solved by technology

[0006] (1) The dipole time series of different types of motor imagery tasks obtained by solving the EEG inverse problem are chaotic, and there are few rules to follow, especially in the case of complex ipsilateral limb motor imagery, the corresponding cortical activation areas may overlap, or They are very close to each other, and only analyzing the time series of dipole and dipole moments in the time domain does not highlight the rhythm characteristics related to motor imagination, so it is difficult to extract features with better separability;

[0007] (2) The number of dipoles obtained after brain-derived imaging is huge, and feature extraction for all dipoles will result in redundant feature information, and a large number of features that are not related to the imagination task will be mixed in, which will affect the classification accuracy; When calculating the zero conduction matrix for the electropositive problem, due to the limitation of experimental equipment, fMRI scanning cannot be performed on each subject, and the head model is usually constructed by numerical calculation based on a general template

There is an approximation in the factor value method, which makes it impossible to adaptively obtain the exact correspondence between scalp electrodes and cortical neurons for different subjects when solving the forward problem

Therefore, there are also errors in the estimation of the dipole distribution obtained through the EEG inverse problem, which affects the accuracy of selecting ROI regions based on the neurophysiological partition method (Brodmann partition);

[0008] (3) The analysis of the dipole during the entire motor imagery period or a specific period of time does not consider the time-varying characteristics of the dipole during the motor imagery period, and the dipole changes caused by different subjects and different motor imagery tasks The impact of the difference on the classification recognition rate

Images