Research On Source Imaging And Feature Extraction Of EEG

Electroncephalogram(EEG)feature extraction is one of the critical technologies of braincomputer interface(BCI).Effective research on feature extraction technology is conducive to better communication between the human brain and the outside world.At present,the braincomputer interface technology has received great attention and attention in the field of medical rehabilitation.Its audience is a large number of brain injury patients,such as Stroke.This type of patients' brain function and structures have been reorganized after the occurrence of brain hurts,and its physiological basis is plasticity(Plasticity).This change can be presented by EEG Source Imaging(ESI)technology.The study of brain function counters the study of the EEG features,adding more possibilities for clinical medicine.In this paper,the signals of healthy subjects and stroke patients are compared and analyzed from the points of EEG classification and source imaging,and the results of source imaging are used to guide the classification of EEG.In order to study the difference of signal between healthy person and stroke patient,the research results of this paper are as follows:(1)Based on the distributed source-LORETA algorithm,BESA toolbox was used to analyze the fisting signal of right or left hand of healthy people and patients.The difference of brain activity between healthy people and patients was obtained from the difference of imaging.From the experimental results,it can be seen that compared with healthy people,the brain function of patients in hospital rehabilitation has indeed changed,this change is reflected in the activation area.(2)EEG data of C3 + C4 channels in healthy subjects were extracted by four different features: frequency band energy,continuous wavelet transform,approximation components and detail components in four-layer discrete wavelet transform.The four features are classified by LDA and SVM classifiers,respectively,using 12 rounds of 5-fold cross-validation.The classification results under four characteristics are obtained.The experimental results show that the approximate components of discrete wavelet transform can be well characterized The classification effect.Using this feature extraction method for patient data,the results of 12 rounds of 5-fold cross-validation with LDA / SVM classifier were also obtained.The results showed that the recognition rate was about 20% lower than that in normal subjects.(3)According to the results of the patient's source imaging,that is,brain activation site is inconsistent with health people.In order to improve the accuracy of the classification of patients,the signal of F3 + F4 channel near the anterior motor cortex was selected and compared with that of healthy people.The result was that the recognition rate decreased after the change of channel in healthy people,and increased in the stroke patients.(4)For other patients data,the electrode pair signals of AFF5 h + AFF6 h and FFC3 h + FFC4 h located in the pre-motor cortex(PMC)and the motor cortex(SMA)were respectively selected to compare the results of the signal from C3 + C4 channels.The results show that in the EEG feature extraction,the approximate components in the discrete wavelet transform can be extracted to get a higher classification result.However,patients with reconstructed brain function and structure due to impaired brain function perform far less than healthy subjects in the same feature extraction and classification results.Analyzing the results of EEG source imaging to analyze the active area of the patient's brain area and change the electrode channel of the feature extraction,good classification results can be obtained.Therefore,the research work in this paper has good research and application value.