Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Front-end Replication Dynamic Window (FRDW) for Online Motor Imagery Classification

X. Chen, J. An, H. Wu, S. Li, B. Liu, D. Wu

TL;DR

This work tackles the challenge of fast and accurate online MI decoding in EEG-based BCIs by introducing FRDW, which combines a dynamic test-trial window with front-end replication to extend short trials to the training length $N$. The method also supports training data augmentation and can be paired with Euclidean alignment (EA) for cross-subject MI classification. Across three public MI datasets, three classifiers, and multiple augmentation schemes, FRDW consistently improves information transfer rate (ITR) and enables real-time operation, with further gains when used with EA for cross-subject transfer. The approach is simple, effective, and practical for real-time BCI deployment and competition-level performance.

Abstract

Motor imagery (MI) is a classical paradigm in electroencephalogram (EEG) based brain-computer interfaces (BCIs). Online accurate and fast decoding is very important to its successful applications. This paper proposes a simple yet effective front-end replication dynamic window (FRDW) algorithm for this purpose. Dynamic windows enable the classification based on a test EEG trial shorter than those used in training, improving the decision speed; front-end replication fills a short test EEG trial to the length used in training, improving the classification accuracy. Within-subject and cross-subject online MI classification experiments on three public datasets, with three different classifiers and three different data augmentation approaches, demonstrated that FRDW can significantly increase the information transfer rate in MI decoding. Additionally, FR can also be used in training data augmentation. FRDW helped win national champion of the China BCI Competition in 2022.

Front-end Replication Dynamic Window (FRDW) for Online Motor Imagery Classification

TL;DR

This work tackles the challenge of fast and accurate online MI decoding in EEG-based BCIs by introducing FRDW, which combines a dynamic test-trial window with front-end replication to extend short trials to the training length . The method also supports training data augmentation and can be paired with Euclidean alignment (EA) for cross-subject MI classification. Across three public MI datasets, three classifiers, and multiple augmentation schemes, FRDW consistently improves information transfer rate (ITR) and enables real-time operation, with further gains when used with EA for cross-subject transfer. The approach is simple, effective, and practical for real-time BCI deployment and competition-level performance.

Abstract

Motor imagery (MI) is a classical paradigm in electroencephalogram (EEG) based brain-computer interfaces (BCIs). Online accurate and fast decoding is very important to its successful applications. This paper proposes a simple yet effective front-end replication dynamic window (FRDW) algorithm for this purpose. Dynamic windows enable the classification based on a test EEG trial shorter than those used in training, improving the decision speed; front-end replication fills a short test EEG trial to the length used in training, improving the classification accuracy. Within-subject and cross-subject online MI classification experiments on three public datasets, with three different classifiers and three different data augmentation approaches, demonstrated that FRDW can significantly increase the information transfer rate in MI decoding. Additionally, FR can also be used in training data augmentation. FRDW helped win national champion of the China BCI Competition in 2022.

Paper Structure

This paper contains 22 sections, 5 equations, 2 figures, 14 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Offline MI Classification
  4. Online MI Classification
  5. Dynamic Window for EEG Classification
  6. Methodology
  7. Problem Setting
  8. FRDW for Online Within-Subject MI Classification
  9. FRDW with EA for Online Cross-Subject MI Classification
  10. Experiments
  11. Dataset and Preprocessing
  12. Data Augmentation
  13. Classifiers
  14. Performance Metric
  15. Experimental Settings
  16. ...and 7 more sections

Figures (2)

  • Figure 1: Relationship between the ITR and classification accuracy for different trial length $T$.
  • Figure 2: Sensitivity analysis on MI1 for within-subject classification. (a)-(c) ITR w.r.t. different $\underline{L}$ ($\tau=0.6$) on EEGNet, CSP+Transformer and CSP+SVM, respectively; (d)-(f) ITR w.r.t. different $\tau$ ($\underline{L}=60$) on EEGNet, CSP+Transformer and CSP+SVM, respectively.