Development Of Emotion Recognition And Evaluation System Based On Photoplethysmography

In recent years,the research of emotion recognition has increased significantly,which can not only excavate the existence of human mental disease,but also have a broad prospect in the application of human-computer interaction and intelligent transportation.Emotion recognition is the processing and analysis of signals to identify people’s emotion.Signals include both nonphysiological and physiological signals.Non-physiological signals contain facial expressions,speech and so on.Since human beings can control their behavior through subjective consciousness and conceal their real emotions,emotion recognition based on non-physiological signals cannot accurately reflect the real emotion.Instead,physiological signals,which mainly come from the central nervous system and the autonomic nervous system,are completely independent of human control and also indirectly reflect the situation of cardiovascular functional activities.Analysis based on physiological signals can improve the impact of subjective consciousness,and can also be used to analyze the mental state through the cardiovascular system.Considering the accuracy,portability and practicability of the identification equipment,the following research work is carried out in this paper:Firstly,the emotion classification algorithm based on Photoplethysmography was conducted.At the beginning,a video-based experiment of eliciting emotions was designed,and PPG signals from multiple data sets were collected as the input of the classification algorithm.Next,filtering,analysis of wave features and automatic detection of feature points were realized.Then,ten different methods(K-Nearest Neighbor,Support Vector Machine,Random Forest,AdaBoost,Gradient Boosting Decision Tree,one-dimensional Convolutional Neural Network,methods which combined network with three different classifiers and Siamese Neural Network)were implemented,the results of classification were compared.Siamese Neural Network is divided into feature extraction and decision net.The step of extracting features consists of two independent networks with the same weight,and two signal sequences form a pair of vectors through two networks respectively.The decision net analyzed whether the two vectors belong to the same emotion according to the distance measurement.The results show that the Siamese Neural Network has good performance,the classification accuracy of four categories can reach 98%,and the recognition time of a single sample is less than 0.5 s.Secondly,the relationship between heart rate variability and mental stress was investigated.According to the physiological significance of heart rate variability,autonomic nervous system,and Photoplethysmography signals,the feasibility of evaluating mental status by heart rate variability was analyzed.The time-domain,frequency-domain and nonlinear parameters of heart rate variability were calculated by collecting long-term signals in different states.The change of parameters between different states was explored,and the evaluation method of fatigue index and other indexes was given,to obtain the report of mental state analysis.Thirdly,the emotion recognition and mental state evaluation system based on Django framework was designed and built.According to the application scenario,the requirements were analyzed,the technical route was determined,and the specific functional modules were designed.It mainly realized the collection of patient information and PPG signal files,embedding recognition and evaluation algorithms,calculating,saving and displaying results,and expanded the functions of report printing and database management.In all,this paper carried out classification and evaluation of emotion based on pulse wave,which can be applied to diagnosis and prevention of mental health,and is expected to be further popularized in stress analysis and cardiovascular status and other fields.