Research On Emotion Classification Methods Based On Neural Network And EEG Signals

Emotion is a complex physiological and psychological phenomenon,which consists of cognitive and logical responses to various situations.Affective computing is computing that originates from emotions and can affect emotions related to emotions.It is an interdisciplinary field that covers computer science,psychology,and cognitive science.As one of the important components of emotional computing,emotion recognition refers to the use of artificial intelligence and other technologies to analyze the physiological or non-physiological signals of an individual to identify the emotional state of the individual.It can not only enhance the interactive experience between man and machine,but also help treat mental disorders such as mental disorders.Among the signals commonly used in emotion recognition,non-physio logical signals include language,facial expressions,body movements and other external manifestations,and physiological signals include ECG,skin electricity,brain electricity,etc.Among them,due to the spontaneous,non-invasive,non-radio active and other advantages of EEG signal,emotion recognition technology based on EEG signal is gradually becoming a hot research direction.In recent years,research on emotion recognition based on deep learning algorithms for EEG and other physiological signals has achieved some remarkable results.However,there are still some problems that need to be solved in this field,such as excessive reliance on time-consuming and complicated preprocessing process and feature engineering,excessive attention to the time-frequency characteristics of the signal and ignoring the spatial topology.In view of the common deficiencies in the current research,this article conducts related research on emotion recognition on the public data set DEAP.The work content mainly includes the following three parts:(1)A simple and efficient preprocessing method for EEG signals is proposed.In this paper,the baseline signal is considered in the research of signal preprocessing.By removing the interference of irrelevant factors in the experimental signal,the signal can better represent the emotional state of the person.Then the LSTM-CNN model,which combines temporal and spatial features,is used to classify emotions.The LSTM network is used to process the information that emotions change over time.The CNN network can mine the correlation between channels,and finally merge the outputs of the two networks to obtain the result of emotion recognition.Experimental results show that the accuracy of this method in the Valence and Arousal dimensions reach 90.69%and 91.31%,respectively,which is better than similar emotion recognition methods based on machine learning and deep learning.In addition,the results of comparative experiments on baseline signal calibration show that the classification accuracy of the model trained using baseline signal calibration data is higher.(2)A new method of expressing spatio-temporal features of EEG signals is proposed.First,according to the original electrode position,the original EEG signal represented by the one-dimensional data vector sequence is converted into two-dimensional EEG frames,and then the two-dimensional frame sequence is connected to the time axis to represent the three-dimensional EEG flow.The three-dimensional EEG flow is used as the input of the 3D CNN emotion classification model based on temporal and spatial features.The exerimental results show that the original EEG signal after three-dimensional reconstruction can be effectively combined with the 3D CNN model.This model achieved 95.16%and 95.57%classification accuracy in the two classifications on the Valence and Arousal dimensions,respectively.In order to further confirm the temporal and spatial validity of this model and its performance superior to manual feature extraction,five models of freely combined input data types and CNN dimensions are used as comparative experimental objects.The experimental results show that the 3D EEG stream proposed by this model shows better performance.(3)The design concept of emotion recognition system based on deep learning and EEG signals is proposed.As an emotion recognition platform,the system realizes emotion recognition and data management based on EEG signals.The main function of the system is to integrate a 3D CNN-based emotion recognition model,detect the emotional state according to the EEG signal data of the subject uploaded by the user,and output the detection result for display and printing.The system can effectively improve the work efficiency of doctors and increase the friendliness of human-computer interaction.