EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization
Due to the limited perceptual field, convolutional neural networks (CNN) only extract local temporal features and may fail to capture long-term dependencies for EEG decoding. In this paper, we propose a compact Convolutional Transformer, named EEG Conformer, to encapsulate local and global features...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Book |
| Published: |
IEEE,
2023-01-01T00:00:00Z.
|
| Subjects: | |
| Online Access: | Connect to this object online. |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Due to the limited perceptual field, convolutional neural networks (CNN) only extract local temporal features and may fail to capture long-term dependencies for EEG decoding. In this paper, we propose a compact Convolutional Transformer, named EEG Conformer, to encapsulate local and global features in a unified EEG classification framework. Specifically, the convolution module learns the low-level local features throughout the one-dimensional temporal and spatial convolution layers. The self-attention module is straightforwardly connected to extract the global correlation within the local temporal features. Subsequently, the simple classifier module based on fully-connected layers is followed to predict the categories for EEG signals. To enhance interpretability, we also devise a visualization strategy to project the class activation mapping onto the brain topography. Finally, we have conducted extensive experiments to evaluate our method on three public datasets in EEG-based motor imagery and emotion recognition paradigms. The experimental results show that our method achieves state-of-the-art performance and has great potential to be a new baseline for general EEG decoding. The code has been released in <uri>https://github.com/eeyhsong/EEG-Conformer</uri>. |
|---|---|
| Item Description: | 1558-0210 10.1109/TNSRE.2022.3230250 |