Research On EEG Emotion Recognition Based On Hybrid Neural Networ

As emotional intelligence is widely used in the field of modern technology,EEG emotion recognition has become the core research content in the field of emotional intelligence.For EEG emotion recognition,there are still two major challenges: one is how to select appropriate emotional features from EEG signals and effectively use the features to realize emotion prediction;the other is how to establish an efficient emotion recognition model and ensure performance Based on the above,the generalization ability of the model is improved by using the complementary emotional information between different signals.In view of the above problems,based on EEG emotion recognition,this paper mainly studies the following three aspects:1.In order to ensure the integrity of the time-space-frequency information of EEG signals,a four-dimensional feature input structure is designed,and a method based on Emotion recognition models for recurrent neural networks and convolutional neural networks.Aiming at the degradation problem of deep neural network,the convolutional neural network is improved by using the residual structure and parallel convolutional layer;for the problem that the input structure contains more irrelevant information,it is solved by introducing the attention mechanism to accurately extract the key points.Information way,improve recognition efficiency and performance.The experimental results prove that comprehensive utilization of space-time-frequency information can effectively recognize EEG emotions,and the constructed emotion recognition model has a certain degree of advancement.2.For irregular dynamic EEG data,an emotion recognition model based on graph convolution and recurrent neural network is proposed.The EEG data in the brain graph is used as the processing object,combined with the correlation of acquisition channels and the differential entropy of different frequency bands Feature Fusion Frequency-Spatial Information.In the feature extraction stage,the current state of the input structure is connected with the historical state using a bidirectional recurrent neural network,and then the time dependence of continuous EEG signals is captured,and the extended selfattention mechanism is introduced to sort the importance of the extracted features.Take full account of historical information.3.Aiming at the heterogeneity and complementarity among different physiological signals,a multimodal signal emotion recognition model combining graph transformation network,graph convolution and recurrent neural network is proposed.In order to utilize the threedimensional information of space-time and frequency at the same time,a dual-stream graph sequence containing space-time and frequency-space information is designed;in order to effectively integrate important emotional information of multi-physiological data,a graph transformation network is used to realize multi-modal data heterogeneity modeling,graph volume The product determines the correlation of multimodal signals,and extracts temporal,spatial and frequency features through a series of gated recurrent units.Experiments have shown that using the correlation and heterogeneity between different physiological signals can effectively improve the performance of emotion recognition.