Research On EEG Emotion Recognition Algorithm Based On Spatio-temporal Convolutional Network Model

In recent years,artificial intelligence has been a hot topic of global research,and the pursuit of the best human-computer interaction experience is one of the starting points of artificial intelligence.The purpose of human-computer interaction is to make computers more humane,and emotion recognition is closely related to human-computer emotional interaction.Emotion recognition technology is one of the key technologies in the fields of artificial intelligence,psychology,and medical treatment,and has important application value.Because electroencephalogram(EEG)has the advantages of convenient extraction and not easy to disguise,EEG emotion recognition has become a hot topic in various fields around the world.EEG emotion recognition models are mainly divided into two categories: traditional machine learning based EEG emotion recognition algorithms and deep learning based EEG emotion recognition algorithms.Traditional machine learning is suitable for small sample training,and the hidden information of EEG cannot be obtained.Although deep learning has a long training time,it is more suitable for a large amount of EEG data and can fully mine the temporal and spatial distribution information of EEG signals.Therefore,this paper uses the advantages of convolutional neural network,empirical mode decomposition and attention mechanism to propose two spatio-temporal convolutional network models and apply them to EEG emotion recognition,and obtain good emotion recognition results.In summary,the main contents of the proposed spatio-temporal convolutional network models are as follows:(1)Subject-independent emotion recognition of EEG based on dynamic empirical convolutional neural networkAiming at the problem of cross-subject emotion recognition,this paper proposes a cross-subject emotion recognition algorithm based on dynamic empirical convolutional neural network(DECNN).Combining the advantages of empirical mode decomposition and differential entropy(DE),this paper proposes a dynamic differential entropy(DDE)feature extraction algorithm.Then,a two-dimensional convolutional neural network is used to classify the DDE features.Finally,this paper uses the data of 15 subjects from the SJTU Emotion EEG Dataset(SEED)and leave-one-subject-cross-validation to evaluate the performance of the model,and gives the comparison results of 5 profiles of electrode placements.The experimental results show that the classification accuracy of the model is good,the best electrode configuration is 20 electrodes,and its average accuracy rate is 97.56%,which has guiding significance for the development of wearable brain electrical equipment.(2)Emotion recognition of EEG based on 3D convolutional attention neural networkAiming at the EEG differences of different emotions,this paper proposes an EEG emotion recognition model based on 3D convolution attention neural network(3DCANN).The model is composed of a spatio-temporal feature extraction module and an EEG channel attention weight learning module,which can well extract the dynamic connection between the multi-channel EEG signals and the internal spatial relationship of the multi-channel EEG signals in a continuous time period.Finally,the spatiotemporal features and the weights of dual attention learning are fused,and finally input into the Softmax classifier for emotion classification.This paper designs three different experiments based on the data of 15 subjects in the SEED data set to comprehensively evaluate the effectiveness of the algorithm.Experimental results show that the algorithm has good performance in EEG emotion recognition.The average accuracy rates of subject-independent,subject-dependent,and single-subject experiments are 96.37%,97.35%,and 97.04%,respectively.In addition,this paper obtained 27 key electrodes through the distribution of attention weights to explore the relationship between the brain and emotions.