Development Of Emotional State Detection System Based On EEG

The emotions play an important role in the perception of external events or situations in people’s daily lives.The emotion recognition has attracted attention in various fields with its broad application prospects,and the research on emotion recognition has become a representative field: emotion computing.There are many ways to recognize emotions,such as facial expressions,voice tone,body movements and other non-physiological signals,as well as physiological signals such as electroencephalogram and electrocardiogram.Among them,the physiological signals have become an important way of existing emotion recognition due to their involuntary and difficult to disguise characteristics.As the most widely used physiological signal in the research field of emotion recognition,EEG signal has the advantages of rapidness and non-invasiveness.The main task of emotional computing is to try to establish a regular and fixed relationship between emotional changes and physiological signals from the perspective of various types of signals,emotional features and classifiers.With the rapid development of deep learning,the emotion computing methods are also advancing with the times.From traditional classification algorithms to deep learning,emotion recognition rates are getting higher and higher.Following this,there are more and more derivative products related to emotion recognition.This thesis develops a We Chat applet based on EEG signal emotion recognition,and specifically does the following work:First,thesis designs and implements an emotion-based EEG signal acquisition experiment,intercepting 40 short vibrato videos of 8-20 seconds each,and 6 film and television dramas in the past five years,each of about 2 minutes,each subject After watching a short video,choose the emotion of the video material.There are two options:positive and negative.A total of 10 subjects’ emotional EEG data were collected and an emotional database was established.Then,thesis uses the recursive structure of long and short-term memory network in time and combines the time-frequency characteristics of EEG to identify emotions.The data sets DEAP,SEED and the data collected in thesis are respectively verified and compared with traditional classification algorithms such as the accuracy of K-nearest algorithm,support vector machine,and Bayesian classification algorithm has been improved.The classification accuracy of LSTM in the DEAP dataset is 65.00% and 71.16%,respectively.The classification accuracy of the higher SVM in the algorithm is 60.35% and64.13%.In the SEED data set,the accuracy of the two classifications is 78.58%.It is higher than 73.50% of SVM,and the accuracy of two classifications on the data set collected in this thesis is 79.91%.The highest recognition rate in the traditional method is the support vector machine,and its accuracy rate is 74.34%.Finally,thesis develops an EEG-based emotion detection We Chat applet based on the above method.The development of the client is based on the We Chat applet developer tool and Tencent Cloud server.The user uploads emotion-related EEG signals to the client to obtain emotions.Recognition results.The client terminal includes functions such as registration,login,uploading signals,emotion detection,records and popular science tweets.The emotion detection module is based on the LSTM model of this article,and uses Django technology to install it as an interface for the front end of the applet.