An Approach for EEG Data Augmentation Based on Deep Convolutional Generative Adversarial Network

High-quality Electroencephalogram (EEG) signals are crucial for BCI applications. In practice, processing and analysis of EEG are often hindered due to either data scarcity or data imbalance, which would lower the performance of EEG-based classification. In this work, we implemented a data augmentation approach based on Deep Convolutional Generative Adversarial Network (DCGAN), which employed a generator and a discriminator to compete with each other to gain high-quality data for augmentation. In the approach, one-dimensional (ID) convolution layers were used by the discriminator to extract time features of raw signals, and ID transposed convolutional layers were adopted by the generator to synthesize new data. The quality of synthesized data was analyzed in time domain, and as well in both frequency and time-frequency domains by using Fast Fourier Transform and Continuous Wavelet Transform. The experimental results showed that the synthesized data have very similar temporal and spectral features to the raw data, which confirmed the effectiveness of our proposed method in synthesizing useable data from limited original signals with a concise structure and fewer network parameters. This work provides a possible deep learning method for EEG data augmentation with a lot of application potentials in BCI areas.