Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EEG-DCNet: A Fast and Accurate MI-EEG Dilated CNN Classification Method

Wei Peng, Kang Liu, Jiaxi Shi, Jianchen Hu

TL;DR

The experimental results show that EEG-DCNet outperforms existing state-of-the-art (SOTA) approaches in terms of classification accuracy and Kappa scores and since EEG-DCNet requires less number of parameters, the training efficiency and memory consumption are also improved.

Abstract

The electroencephalography (EEG)-based motor imagery (MI) classification is a critical and challenging task in brain-computer interface (BCI) technology, which plays a significant role in assisting patients with functional impairments to regain mobility. We present a novel multi-scale atrous convolutional neural network (CNN) model called EEG-dilated convolution network (DCNet) to enhance the accuracy and efficiency of the EEG-based MI classification tasks. We incorporate the $1\times1$ convolutional layer and utilize the multi-branch parallel atrous convolutional architecture in EEG-DCNet to capture the highly nonlinear characteristics and multi-scale features of the EEG signals. Moreover, we utilize the sliding window to enhance the temporal consistency and utilize the attension mechanism to improve the accuracy of recognizing user intentions. The experimental results (via the BCI-IV-2a ,BCI-IV-2b and the High-Gamma datasets) show that EEG-DCNet outperforms existing state-of-the-art (SOTA) approaches in terms of classification accuracy and Kappa scores. Furthermore, since EEG-DCNet requires less number of parameters, the training efficiency and memory consumption are also improved. The experiment code is open-sourced at \href{https://github.com/Kanyooo/EEG-DCNet}{here}.

EEG-DCNet: A Fast and Accurate MI-EEG Dilated CNN Classification Method

TL;DR

The experimental results show that EEG-DCNet outperforms existing state-of-the-art (SOTA) approaches in terms of classification accuracy and Kappa scores and since EEG-DCNet requires less number of parameters, the training efficiency and memory consumption are also improved.

Abstract

The electroencephalography (EEG)-based motor imagery (MI) classification is a critical and challenging task in brain-computer interface (BCI) technology, which plays a significant role in assisting patients with functional impairments to regain mobility. We present a novel multi-scale atrous convolutional neural network (CNN) model called EEG-dilated convolution network (DCNet) to enhance the accuracy and efficiency of the EEG-based MI classification tasks. We incorporate the convolutional layer and utilize the multi-branch parallel atrous convolutional architecture in EEG-DCNet to capture the highly nonlinear characteristics and multi-scale features of the EEG signals. Moreover, we utilize the sliding window to enhance the temporal consistency and utilize the attension mechanism to improve the accuracy of recognizing user intentions. The experimental results (via the BCI-IV-2a ,BCI-IV-2b and the High-Gamma datasets) show that EEG-DCNet outperforms existing state-of-the-art (SOTA) approaches in terms of classification accuracy and Kappa scores. Furthermore, since EEG-DCNet requires less number of parameters, the training efficiency and memory consumption are also improved. The experiment code is open-sourced at \href{https://github.com/Kanyooo/EEG-DCNet}{here}.

Paper Structure

This paper contains 14 sections, 6 equations, 4 figures, 10 tables.

Table of Contents

  1. Introduction
  2. EEG-DCNet Architecture
  3. Data Preprocessing
  4. Convolutional Block
  5. Multi-Branch Parallel Atrous Convolution Block
  6. Sliding Window and SE Attention Block
  7. Experiment and Result Analysis
  8. Dataset
  9. Evaluation Metrics
  10. Comparison of number of sliding windows
  11. Comparison of Different Attention Mechanisms
  12. Ablation Analysis
  13. Baseline Comparison
  14. Conclusion

Figures (4)

  • Figure 1: The architecture of EEG-DCNet. The architecture of EEG-DCNet consists of several key components: initial data preprocessing is followed by a convolutional block that extracts spatial and frequency features. An average pooling and dropout layer then perform dimensionality reduction and regularization. Next, a multi-branch parallel atrous convolution block captures multi-scale features at varying dilation rates. This is followed by a sliding window and SE attention module to enhance local feature representations. Finally, fully connected layers produce the output.
  • Figure 2: The architecture of CV block.
  • Figure 3: The architecture of AsrtoConv.
  • Figure 4: The architecture of EEG-DCNet.