MI-EEG Temporal Information Learning Based on One-Dimensional Convolutional Neural Network

The Brain-Computer Interface (BCI) is a special communication system that does not rely on the nerve and muscle tissue around the human body. it realizes the direct communication between the brain and the computer. The key to the realization of the BCI system is reasonable extraction and analysis of the characteristics for electroencephalogram (EEG) signals. Although deep learning has made initial progress in this field, only a few works have discussed the characteristics of EEG signals learned by deep learning. In this paper, an end-to-end network model for deep learning is designed to extract the time domain information of MI EEG signals. The model uses a one-dimensional convolution kernel to learn the time-domain information in the MI-EEG signal. Furthermore, we analyzed the frequency-dependent characteristics of the convolution kernel. We use the BCI competition dataset as a standard to measure the performance of this method. The experimental results show that the neural network tends to learn information in a specific frequency band, and each person's specific frequency band is different. The classification result of the motor imagery obtained by the network is better than the competition's winner.