TSception: Capturing Temporal Dynamics and Spatial Asymmetry from EEG for Emotion Recognition

The high temporal resolution and the asymmetric spatial activations are essential attributes of electroencephalogram (EEG) underlying emotional processes in brain. To learn dynamics asymmetry EEG towards accurate generalized emotion recognition, we propose TSception, a multi-scale convolutional neural network that can classify emotions from EEG. TSception consists dynamic temporal, spatial, high-level fusion layers, which discriminative representations time channel dimensions simultaneously. layer 1D kernels whose lengths related to sampling rate EEG, learns frequency takes advantage patterns for emotion, learning global hemisphere representations. learned will be fused by layer. Using more cross-validation settings, proposed method is evaluated on two publicly available datasets DEAP MAHNOB-HCI. performance compared with prior reported methods such as SVM, KNN, FBFgMDM, FBTSC, Unsupervised learning, DeepConvNet, ShallowConvNet, EEGNet. achieves higher classification accuracies F1 scores than other most experiments. codes at https://github.com/yi-ding-cs/TSception