Deep convolutional neural network for decoding motor imagery based brain computer interface

Abstract

Deep convolutional neural networks (CNNs) have made tremendous development in the field of image recognition and natural language processing. However, there is still a lack of knowledge of using CNN models to decode motor imagery based Brain Computer Interface (BCI). This paper presents a method applying CNN to analyze the EEG signals, which are produced by left and right hand motor imagery tasks. EEG signals are transferred into time-frequency images using Short-Time Fourier Transform (STFT) and then those images are fed as input of the network for classification. A comparison is made to verify the performance of three different activation functions during the network's learning procedure. They are rectified linear unit (ReLU), exponential linear unit (ELU) and the newly proposed scaled exponential linear unit (SELU). The results show that our method using CNN model can achieve better accuracy than the conventional method and SELU function shows superior ability for the network to convergence.