Emotional Classification Based On EEG

Emotion is the reaction of people when they are stimulated by external stimulation.It is not only the supervisory feeling produced by the human heart,but also the objective physiological phenomenon produced by the human body.It is a coordinated reaction of language,physiology,behavior and neural mechanism.Emotions play an important role in interpersonal communication in our daily life.With the development of science and technology,the technology of human-computer interaction has gradually matured and received more and more attention.If the machine can automatically recognize people's emotions in the process of human-computer interaction,the interactivity process will be more intelligent and greatly enhance the sense of experience.At the same time,the recognition of emotions can assist in the treatment of mental and mental diseases,and has a wide range of uses in medical public security and other fields.This paper studies the algorithm of emotion classification based on EEG signals.The main work is as follows:1.We use a portable dry electrode device to construct an emotional dataset containing 14 people's EEG signals.The categories of EEG signals are:positive,neutral and negative.Emotion is induced by giving participants a way to watch the movie.This data set can be used for experiments,training,and testing of emotional algorithms.2.This paper proposes an emotion classification algorithm based on EEG signals.The algorithm mainly consists of three steps:the first step is to filter out the noise,and the amplitude and wavelet transform are used to remove noise from both the time domain and the fr-equency domain;the second step is feature extraction,and the differential entropy of the five important bands of the EEG signal are calculated which is converted into a five-channel EEG feature map.The third step is to train the classifier.The classifier is mainly composed of three parts:CNN,LSTM and Attention Model.The CNN is responsible for learning the characteristics of the image frequency domain and the spatial domain.The LSTM is responsible for learning the time domain relationship of the image.According to the idea of the attention module,the most important segment of the sequence and the output result of the LSTM are selected and connected as the final output vector.At the same time,the model training method based on transfer learning is used.Firstly,the non-time domain part of the network is trained,and the parameter settings of the network are saved.When the time domain network is trained,the parameters are initialized to reduce the time cost of network training.3.The effectiveness of the proposed classification was verified in the public EEG emotion dataset SEED.At the same time,another method for dealing with multi-channel EEG features was selected.The equidistant azimuthal projection is used to project the three-position coordinates between the EEG electrode points into a two-dimensional plane,and the feature values are filled into the corresponding electrode positions in the form of colors,which is similar to the input when processing the single-channel EEG signals.The classification accuracy rate is also obtained in the data set SEED,which proves the rationality and effectiveness of the network design.