Study On Pattern Identification Method In Motor Imagery Brain-Computer Interface By FNIRS

With the development of cognitive neuroscience technology,the objective measurement of various physiological parameters can be realized.As a promising and non-invasive technique,functional near-infrared spectroscopy?fNIRS?can easily detect the function of the brain.It has the advantages of high time resolution?strong anti-interference ability?safety and easy implementation.For the extraction of cerebral cortical functional information and revealing the advanced functions of the brain?understanding the brain disease mechanism and clinical brain protection it has important application value.Providing valuable monitoring data for clinical medicine and scientific research.The concept of motor imagery was the exercise activity only takes place in the consciousness,and is not accompanied by any real peripheral nerve and muscle movement.The original purpose of the study was to use it in the military field,later it was expanded to civilian research,which combined motor imagery with physical exercise to improve the level of exercise.In recent years,motor imagery has gradually become an auxiliary treatment for clinical rehabilitation of people with movement disorders.In addition,it also has important value in the monitoring of cerebral oxygen in newborns.This paper mainly considers the characteristics of multi-measurement indexes and high temporal resolution of fNIRS technique.Taking motor imagery as a breakthrough point,this paper studied the brain imaging pattern recognition method based on fNIRS.Mainly in the following sereval aspects to study,and achieved certain results:?1?Firstly,it elaborated the research foundation of BCI,motor imagery and fNIRS.An experimental paradigm of spatial four-direction movement imagination was designed.?2?According to this paradigm,blood oxygen concentration signals of healthy subjects in the task state were collected.The collected signals were mixed with many noises such as respiration?0.4Hz?,heartbeat?1.5Hz?,and blood pressure?0.1Hz?.We performed preprocessing on this signal.First,the original signal should through a 0.01-0.2Hz band-pass filter,and then calculate the blood oxygen concentration based on the Modified Lambert-Beer law.?3?There is still a fatal problem with the signal after the above processing-frequency aliasing of 0.1Hz Mayer wave.In order to solve this problem,we proposed a method of combining EEMD-ICA to remove and compared it with the traditional EMD method.The results show the method that we proposed had better results and the frequency aliasing problem was also solved.?4?Using linear discriminants?LDA?and support vector machines?SVM?method in our four-dimensional motor imagery data set.Considering the limitation of fNIRS technique,we extracted the currently used average feature as an input of the classifier.Calculated the average concentration of(7₂,Hb and O₂sat under 8-14s,8-21s different time windows,for the final classification results,we compared the results of two classifiers.Finally,the optimal activation channel map was produced which combine the brain map of 10 subjects and its matrix mapping according to concentration of(7₂.This study has completed some work and achieved certain results,providing a theoretical basis for the future application of fNIRS-BCI.