Research On Feature Extraction Method Of Intimate Relationship Prediction Based On FNIRS

Relationship is a description of the degree to which individuals influence each other or depend on each other.When the degree of dependence between the two is great,it is called intimacy.The study of intimacy is very important for maintaining various interpersonal relationships in life.At present,most researches on intimate relationships use functional magnetic resonance imaging(fMRI),which proves that the dopamine pathway can regulate the activation of love-related brain regions.However,fMRI has the disadvantages of closed measurement environment and complicated noise interference sources.However,functional near-infrared spectroscopy(fNIRS)technology is not constrained by subjects and can be directly used to measure changes in blood oxygen concentration with high time resolution related to nerve activation.Therefore,this article uses three groups of photos of "friends","lovers",and "strangers" to represent the three types of intimacy,and uses the fNIRS device to measure the brain nerve activity of the subjects when viewing the three types of pictures and analyze the brain neural mechanism.At the same time,it uses shallow machine learning and deep learning models to predict intimacy categories.This paper analyzes the neural mechanism of intimacy through t-test and brain function connection,and explores the response pattern of intimacy expressed on brain imaging.The results show that there are different activation areas when viewing different types of photos.Based on the analysis results of the neural mechanism,different machine learning methods are used to extract the characteristics of the fNIRS signal,and the classification of the intimate relationship is studied.The general linear model(GLM)method is used to extract the ? value,and the complex brain network analysis(CBNA)method is used to extract the measurement indicators of the local nodes of the brain network with significant differences as the feature vectors.The prediction accuracy rate is 98.89%.The use of GLM and CBNA requires a professional technical background.Inorder to automatically extract the characteristics of the intimate relationship category,this article sets up a serial convolutional neural network(CNN)and Long Short Term Memory Network(LSTM)simultaneously capture the temporal and spatial information of fNIRS data to predict the intimacy category based on brain imaging representation.In order to solve the long-term time dependence of LSTM in learning fNIRS time series signals,an attention mechanism(Attention mechanism)is introduced to extract the salient features of key information about intimacy in fNIRS signal fragments.The prediction accuracy rate is 97.39%.Although the prediction accuracy of this method is not as good as the prediction results of the shallow machine learning method based on CBNA feature extraction,it can automatically extract features and can effectively learn intimate relationships.