The data scarcity problem in emotion recognition from electroencephalography (EEG) leads to difficulty building an affective model with high accuracy using machine learning algorithms or deep neural networks. Inspired by emerging generative models, we propose three methods for augmenting EEG training enhance the performance of models.Our proposed are based on two variational autoencoder (VAE) and adversarial network (GAN), augmentation ways, full partial usage strategies. For strategy, all generated augmented dataset without judging quality data, while usage, only high-quality selected appended dataset. These called conditional Wasserstein GAN (cWGAN), selective VAE (sVAE), WGAN (sWGAN).To evaluate effectiveness these methods, perform a systematic experimental study public datasets recognition, namely, SEED DEAP. We first generate realistic-like forms: power spectral density differential entropy. Then, augment original different number data. Finally, train support vector machines networks shortcut layers build models datasets. results demonstrate that our outperform existing approaches such as VAE, Gaussian noise, rotational augmentation. also observe should be less than 10 times achieve best performance.The produced sWGAN method significantly EEG-based models.

Data augmentation for enhancing EEG-based emotion recognition with deep generative models