A Novel Neural Network Model based on Cerebral Hemispheric Asymmetry for EEG Emotion Recognition

Abstract

In this paper, we propose a novel neural network model, called bi-hemispheres domain adversarial neural network (BiDANN), for EEG emotion recognition. BiDANN is motivated by the neuroscience findings, i.e., the emotional brain's asymmetries between left and right hemispheres. The basic idea of BiDANN is to map the EEG feature data of both left and right hemispheres into discriminative feature spaces separately, in which the data representations can be classified easily. For further precisely predicting the class labels of testing data, we narrow the distribution shift between training and testing data by using a global and two local domain discriminators, which work adversarially to the classifier to encourage domain-invariant data representations to emerge. After that, the learned classifier from labeled training data can be applied to unlabeled testing data naturally. We conduct two experiments to verify the performance of our BiDANN model on SEED database. The experimental results show that the proposed model achieves the state-of-the-art performance.

From Regional to Global Brain: A Novel Hierarchical Spatial-Temporal Neural Network Model for EEG Emotion Recognition

“…In this experiment, we adopt the similar subject-dependent EEG emotion recognition protocol used in [7] and [15] to evaluate the proposed method, where both training and testing data come from the same subject but different EEG trials. Specifically, we choose 9 trials of EEG signals in every session to serve as training data set and use the other 6 trials from the same session as testing data.…”

“…• Three baseline methods: Support Vector Machine (SVM) [30], Random Forest (RF) [31], and Canonical Correlation Analysis (CCA) [32]; • Two subspace learning methods: Group Sparse Canonical Correlation Analysis (GSCCA) [8] and Graph Regularized Sparse Linear Regression (GRSLR) [9]; • Five deep learning methods: Deep Belief Networks (DBN) [7], Graph Convolutional Neural Networks (GCNN) [33], Dynamical Graph Convolutional Neural Networks (DGCNN) [14], Domain Adversarial Neural Network (DANN) [34], and Bi-hemisphere Domain Adversarial Neural Network (BiDANN) [15]. Table II shows the experimental results of the various methods, from which we can see that the proposed R2G-STNN method achieves the recognition accuracy as high as 93.38%, which is the best recognition result among the various recognition methods.…”

“…In this experiment, we will investigate the subjectindependent EEG emotion recognition problem, in which the training EEG data samples and the testing ones come from different subjects [15] [46]. To this end, we adopt leaveone-subject-out (LOSO) cross validation strategy to conduct the experiment, in which we circularly take one subject's EEG signals as testing data and the EEG signals of all the other subjects as training data.…”

“…• Two baseline methods: SVM [30] and kernel principal component analysis (KPCA) [39]; • Nine transfer subspace learning methods: Kullback-Leibler importance estimation procedure (KLIEP) [36], unconstrained least-squares importance fitting (UL-SIF) [37], selective transfer machine (STM) [38], transfer component analysis (TCA) [40], transfer kernel learning (TKL) [41], subspace alignment (SA) [42], geodesic flow kernel (GFK) [43], transductive SVM (T-SVM) [44], and transductive parameter transfer (TPT) [45]; • Three recent deep learning methods: DGCNN [14], DANN [34] and BiDANN [15]. The experimental results of the various methods are shown in Table IV, from which we can again see that R2G-STNN achieves the state-of-the-art performance among the various methods.…”

“…Song et al [14] constructed a graph relation based on multichannel EEG data and then performed graph convolution on it to extract feature for classification. Li et al [15] considered the domain shift for EEG data, and utilized the difference between two brain hemispheres to decrease this shift and achieved the state-of-the-art performance.…”