 The proposed DVJN model generates high-quality artificial samples to improve the performance of deep learning models in EEG-based emotion recognition tasks. This model uses differential entropy features extracted from EEG data combined with spatial and temporal morphological features to create two types of latent variables. These variables are then fed into a decoder to generate artificial samples. The augmented training datasets generated by the proposed method have an average accuracy of 97.21%, which is a 5% improvement over the original dataset. Additionally, the similarity between the generated data and the original data distribution has been proven. This article was authored by Chen Xitian, Yu Liangma, Jared Kamen, and others.