EEG Emotion Recognition Based On Spatial Asymmetric Convolution Network

Emotion is closely related to people’s daily life,and emotion recognition is the key to emotion research.In recent years,considering the difficult camouflage and strong real-time characteristics of EEG signals,emotion recognition based on EEG signals has attracted wide attention of scholars at home and abroad.It can help people better explore the mechanism of emotion generation.However,there are still some problems in current research,such as blind and insufficient feature extraction of emotion recognition of subject-dependent,and large differences in EEG of subject-independent.Therefore,this paper studies the EEG emotion recognition based on spatial asymmetric convolution network.The main research work includes:(1)Aiming at the problems of blind and incomplete EEG feature extraction and the difference of emotional response between the left and right hemispheres of the brain,this paper proposes a subject-dependent EEG emotion recognition model based on Multi-Band Spatial Asymmetry Convolution Neural Network(MB-SACNN).First of all,the EEG signal is processed by frequency division to explore the role of each frequency band and frequency band combination in emotion recognition;After that,the original EEG matrix and the spatial asymmetric EEG matrix are constructed by fully considering the spatial asymmetry of the brain to capture the different responses of the left and right brain to emotional stimuli;Finally,the network structure of double-input and single-output is designed,and the two EEG matrices are used as inputs.The convolution neural network is used to fully mine the time-frequency and spatial characteristics of EEG signals,so as to obtain better classification accuracy.The experimental results show that the accuracy of this model can be 98.07% and 96.61% for arousal and valence on DEAP database respectively,which is superior to other comparison methods,and proves the effectiveness and superiority of the model.(2)In view of the complex distribution of multi-source migration scene data and the insufficient feature extraction of a single structure,a cross-subject EEG emotion recognition model based on Multi-Source Adaptation Spatial Asymmetry Convolution Neural Network(MSA-SACNN)was proposed.First,the common feature extractor is designed to map multiple source and target domain data from the original feature space to a common sharing potential space,and then extract the underlying common feature representation of each domain;Secondly,build a multi-domain specific feature extractor,which extracts data from different source domains into different feature spaces,so as to align the feature distribution of the source domain and the target domain in different feature spaces.Each specific domain feature extractor is a multi-representation domain adaptation network,which can more comprehensively extract the high-level domain invariant features between the two domains;Finally,a domain-specific classifier is designed to minimize the difference between all classifiers and align the output of the domain-specific classifier of the target sample.Verified on the DEAP dataset,the results show that the model has good classification performance in cross-subject EEG emotion recognition.