Classification of Motor Imagery EEG Signals Based on Channel Attention Mechanism

Brain-computer interface (BCI) technology establishes communication between the brain and external devices by decoding EEG signals. BCI technology based on motor imagery (MI) has great application potential. There are many different methods to extract motor intention from electroencephalogram (EEG) based on motor imagery (MI).These methods rely on extracting the unique features of EEG in the process of imaginary movement, which directly affect the performance of neural decoding algorithm of BCI. Convolutional neural network (CNN) shows outstanding advantages in automatic extraction of image features. In this paper, an image representation method based on the EEG is proposed as the input of the network. Then, a CNN and a CNN based on Channel Attention Mechanism (CAM) are built as the classifier, convolution layers and activation functions of different sizes are validated. The performance of the method is evaluated. A CNN framework based on CAM, which contained three convolution layers (3-L) is better than the other state-of-the-art approaches. The accuracy on dataset IV from BCI competition II reaches 72.6%.