Construction Of Emotional And Physical Stress Dataset And Stress Classification Based On Facial Tissue Oxygen Saturation

Stress has an important relationship with human physical and mental health as well as life happiness,and long term stress can cause serious harm to general health.Stress can be divided into emotional stress(ES)and physical stress(PS)depending on different stressor sources.These two kinds of stress share some similar physiological characteristics,and can be confused when the stimulus content is unknown.Most previous stress detection approaches focused on one stress type,using cumbersome contact methods to obtain the signal.To address the above problems,in this thesis,the physiological signal of tissue blood oxygen saturation(StO2)was extracted from facial tissue based on hyperspectral imaging technology(HSI)to detect stress remotely.The main work and contributions of this study is as follows:(1)A StO2-based stress dataset was built.We designed a suitable stress inducing method for experimental environments,and collected hyperspectral images for ES,PS,and corresponding baseline from 42 subjects.The original facial StO2 image was generated by extracting the data of 33 wavelength bands from the hyperspectral face image.For convenience,we registered all the original StO2 images for different genders and face sizes to the standard human face,providing standard facial StO2 data.This dataset was made public,which is the first publicly available StO2-based psychophysiological stress dataset.(2)The changes of facial StO2 under the two different stress states were analyzed.Under ES,facial StO2 of all subjects increased overall.Under PS,facial StO2 decreased overall,but the changes in facial StO2 varied widely among subj ects.Finally,we provided the StO2 t-maps of different types of stress for the first time.(3)Optimal StO2 feature combination to classify these two stress types were extracted depending on StO2 variation patterns.Average StO2 differences for seven regions of interest(ROIs)before and after stress stimulation were extracted as features,using an exhaustive method to list all possible feature combinations then input them into a support vector machine(SVM)for classification.Experimental results confirmed that feature combination left cheek,chin,and the region between the eyebrows achieved highest accuracy(95.56%).We clarified that optimal feature combinations arise from feature combination effects.Finally,we compared our StO2-based stress classification method with other methods,the results showed that good performance can be achieved using only a small number of StO2 features and a simple SVM classifier.The results of this study verify that facial StO2 obtained by HSI is an important physiological signal that can be used to distinguish different stress states,providing a new effective input model for stress classification.