Dual-Encoder VAE-GAN With Spatiotemporal Features for Emotional EEG Data Augmentation

Abstract

The current data scarcity problem in EEG-based emotion recognition tasks leads to difficulty in building high-precision models using existing deep learning methods. To tackle this problem, a dual encoder variational autoencoder-generative adversarial network (DEVAE-GAN) incorporating spatiotemporal features is proposed to generate high-quality artificial samples. First, EEG data for different emotions are preprocessed as differential entropy features under five frequency bands and divided into segments with a 5s time window. Secondly, each feature segment is processed in two forms: the temporal morphology data and the spatial morphology data distributed according to the electrode position. Finally, the proposed dual encoder is trained to extract information from these two features, concatenate the two pieces of information as latent variables, and feed them into the decoder to generate artificial samples. To evaluate the effectiveness, a systematic experimental study was conducted in this work on the SEED dataset. First, the original training dataset is augmented with different numbers of generated samples; then, the augmented training datasets are used to train the deep neural network to construct the sentiment model. The results show that the augmented datasets generated by the proposed method have an average accuracy of 97.21% on all subjects, which is a 5% improvement compared to the original dataset, and the similarity between the generated data and the original data distribution is proved. These results demonstrate that our proposed model can effectively learn the distribution of raw data to generate high-quality artificial samples, which can effectively train a high-precision affective model.

Cross-Sensory EEG Emotion Recognition with Filter Bank Riemannian Feature and Adversarial Domain Adaptation

“…Baseline correction was conducted within the initial 1000 ms before stimulation onset. The primary focus of the investigation was the EEG data spanning 0–30 s during stimulation, utilizing a 5 s window [ 37 , 38 , 39 ]. A total of 7200 samples were analyzed, comprising 60 samples per participant for each of the six cross-sensory EEG data types, each sample containing 2500 temporal features derived from 500 × 5 data points.…”

A Data Augmentation Method for Motor Imagery EEG Signals Based on DCGAN-GP Network

“…Abdelfattah et al conducted one of the earliest studies to augment MI signals using GANs, where they introduced a Recursive Generative Adversarial Network (RGAN) model to generate synthetic EEG data to increase the dataset size [35]. Compared to Autoencoders (AE) and Variational Autoencoders (VAE) and their improvements [36], RGANs utilize a cyclic structure to model time-series data, which can better capture the temporal correlations in EEG data, thereby improving classification accuracy. However, an RGAN may face the issue of vanishing or exploding gradients when dealing with long sequence data, leading to training instability.…”

Self-Supervised Data-Driven Approach Defines Pathological High-Frequency Oscillations in Human

“…Autoencoders, such as variational autoencoders (VAEs), 25 is a DL framework based on a self-supervised training paradigm, which can provide such a capability and has been proven to be successful in many natural language processing, 26 computer vision tasks, 27 medical data, including EEG analysis. 28,29 VAEs are generative models that learn efficient low-dimensional representations (latent encodings) of unlabeled data by requiring that the original high-dimensional data be reconstructed from their latent codes. It is particularly effective in discovering clusters in the latent space, where each cluster corresponds to a different mechanism that led to the generation of data in it.…”