Research On Pressure Recognition Based On Non-contact Heart Rate Variability

Heart Rate Variability(HRV)is a physiological signal that reflects the activity of the human autonomic nervous system and can be used for emotion recognition.At present,there are many studies on HRV-based emotion recognition such as happiness,sadness,fear,etc.,but there are few studies on the relationship between HRV characteristics and stressful emotions.In daily life,it is very important to analyze and monitor people’s stress state without interference.Therefore,this paper focuses on the research on stress recognition based on non-contact HRV features.Due to stress-induced difficulties,there are no publicly available datasets dedicated to stressful emotions.The DEAP dataset is a multi-channel database established by Queen Mary University of London and others to study human emotional states.By consulting a large number of psychological research literatures,this paper uses the Pleasure-Arousal-Dominance(PAD)three-dimensional emotional model to map the stress emotion to the determined range of arousal(Arousal)and pleasantness(Pleasure)on the PAD model.The physiological signals that meet the conditions are screened from the DEAP database,and the samples with abnormal HRV characteristics are eliminated by the interquartile range method to form a data set for stress recognition.In order to identify stress based on HRV features,this paper proposes a S-R(Spearman-Relief)hybrid strategy feature screening method based on correlation ranking and contribution ranking on the basis of analyzing the correlation between HRV features and stressful emotions.The Spearman method was used to analyze the correlation of HRV features under stress and calm state respectively,and the correlation ranking vector S of 28 HRV features was calculated,and the Relief algorithm was used to calculate the classification contribution ranking vector R of each HRV feature to stress recognition.The linear summation of vectors determines the weight coefficients of S and R to obtain the rank sum vector W,and then selects the optimal feature subset to construct the pressure identification model.Experiments on the data set selected from DEAP show that using the method proposed in this paper,the accuracy rate of stress recognition based on random forest reaches 75.56%,which is 11.14% higher than that without feature screening,and is higher than that of directly using Relief to screen features.An increase of 5.32%.When HRV features are extracted in a non-contact way based on face videos,HRV features will be affected by factors such as acquisition equipment and environment,which in turn affects the accuracy of pressure recognition.In order to improve the accuracy of pressure recognition based on non-contact HRV features,this paper proposes a hybrid feature screening strategy based on error analysis.This method analyzes non-contact HRV features and synchronization on the basis of piecewise linear interpolation to reduce accuracy errors.For the error of contact HRV feature,the TOPSIS method is used to fuse the feature error factor and SR feature screening to obtain the ESR(Error-Spearman-Relief)feature screening method.The experimental results show that the accuracy of pressure recognition based on the non-contact HRV feature using the method in this paper reaches 70.42%,which is 5.14% different from the pressure recognition accuracy of the contact HRV feature collected synchronously.The rate increased by 18.63%.