Research On Classification And Recognition Of Multi-type Evokes Emotional EEG Signals

Emotion recognition is one of the important research directions in the field of affective computing.Human emotions can be recognized by various physiological signals such as speech,facial expressions,EEG,and ECG.EEG signals can reflect various thinking states of the brain and the neural activity of the brain when representing emotional changes,so it can be used as the object of emotion recognition.This paper mainly expounds its research status from the two perspectives of the induction method and state recognition of emotional EEG signals.basing on the regular performance of EEG signals and emotion-related characteristics,experiments on emotional EEG signals induced by pictures,music and videos were designed,Collect and analyze its response effect,combining different classification algorithms to realize the recognition of positive,neutral and negative emotions.It provides a theoretical basis for building a multi-type material-induced emotional EEG signal recognition model.The emotion-related characteristics of EEG signals under multiple types of evoked materials were explored,and the energy ratio analysis method of emotional EEG signals based on time-domain energy entropy was applied.According to the emotion quantification standard,the subjects selected pictures,music and video test materials that evoked the target emotion,and the average evoked effects of different emotions were clearly distinguished;Analyzing the energy proportion of EEG signals from the perspective of time,it is found that the emotional EEG signals induced by the above three types of materials have a higher proportion of energy in the temporal lobe leads on both sides during the period of obvious feeling.The brain topography and time-spectrogram of EEG signals in positive,negative and neutral emotional states were studied.The results showed that positive emotions showed significant energy increase in both temporal lobe regions of beta and gamma frequency bands,while negative emotions showed energy decline in both beta and gamma frequency bands.The right temporal lobe area of the brain may be more pronounced and intense in energy changes.The PSD features of 5 frequency bands of EEG signals under 3 kinds of emotion-inducing materials were extracted,and the recognition accuracy of EEG signals in three categories of EEG signals under extreme learning machine,support vector machine and random forest classification algorithm was studied.From the perspective of evoked material,the classification effect of music-induced EEG signal is better,the average classification accuracy is 95.09%,91.97%,97%,and the signal shows relatively good feature convergence;The results are discussed from the perspective of classification methods.It is found that the three-classification accuracy of extreme learning machine and support vector machine is about88%.The random forest method has high accurate recognition accuracy and maintains good stability.Explored the application of the deep learning method based on the interleaved group structure convolutional neural network in the three-classification task of emotional EEG signals.The Le Net-5 convolutional neural network model was constructed combined with the lightweight interleaved group structure IGC to classify the EEG PSD feature data,and the average accuracy of the final EEG signal emotion three-classification task was 98.76%.The experimental results show that the CNN-IGC classification model has certain advantages in classification performance and recognition accuracy.