Stress Detection Based On Multi-channel Physiological Signals

As the pace of life accelerates and social competition becomes more and more fierce,people’s psychological stress in their studies,work,and life has become more severe and common.Excessive psychological stress harms the human body in psychology,emotion,cognition,and behavior and causes stress-related diseases.How to accurately detect and intervene in psychological stress through various physiological and behavioral data of the human body has always been a hotspot and focus of research.However,the existing research still has some weaknesses in the construction of data sets and recognition algorithms.In this paper,a multi-modal dataset for psychological stress detection is established,and a multi-channel physiological feature fusion method of psychological stress detection is proposed.The main work of this theme is as follows:(1)Design psychological stress inducing experiments and collect data.Stroop color-word test,Stroop number-size test,rotating letter test,and Kraepelin test are used to induce the subject’s psychological stress.Collect the subject’s pulse and electrical skin activity data,and record the subject’s facial videos while testing.Verify the stress-inducing effect by analyzing the subjective ratings,and finally,create a multi-modal psychological stress detection dataset containing 120 people.(2)Analyze physiological signals and extract features.Use discrete Fourier transform to obtain pulse frequency domain information,compare and analyze the amplitude spectrum,and Poincare scatter map of pulse signal under different ratings.From filtered pulse and electrical skin activity data to extracting time-domain features,frequency-domain features,Non-linear features,and unique features of the signal.(3)Design a psychological stress detection model.Use different machine learning methods to train on feature sets to obtain baseline models for stress detection;design a one-dimensional convolution model,LSTM model,and feature fusion model,and use filtered physiological signals and wavelet decomposition coefficients to training model to verify the validity of models.Through experimental comparison and analysis,the method proposed in this work to use wavelet decomposition coefficients instead of original physiological signals as model input can improve stress detection accuracy.Furthermore,the designed feature-layer fusion model based on multi-channel physiological signals can further improve the performance of stress detection.