Xem trước tài liệu

Đang tải tài liệu...

Thông tin chi tiết tài liệu

Định dạng: PDF
Số trang: 13 trang
Dung lượng: Đang cập nhật

Giới thiệu nội dung

Feature Selection and Blind Source Separation in an EEG-Based Brain-Computer Interface

Author:

David A. Peterson, James N. Knight, Michael J. Kirby, Charles W. Anderson, Michael H. Thaut

Field:

EURASIP Journal on Applied Signal Processing

Document Content:

This paper introduces an EEG-based brain-computer interface (BCI) that aims to directly discriminate between “yes” and “no” responses within a single session, contrasting with typical BCI systems that often require extensive biofeedback training or indirect task mappings. The research explores the effectiveness of signal processing techniques, specifically Blind Source Separation (BSS) and spectral transformations of EEG data, combined with feature selection methods to enhance classification accuracy. By transforming the EEG into a 180-dimensional feature space and employing a modified genetic algorithm (GA) with a support vector machine (SVM) classifier, the study investigates feature subsets that outperform the full feature set and random subsets. The findings suggest that BSS transformations of EEG data, particularly when coupled with feature selection, significantly improve the performance of direct, single-session BCIs by yielding features with stronger discriminative power compared to original EEG signals. This approach offers a promising avenue for developing more efficient and potentially faster BCI systems, especially for individuals with motor impairments.

Table of Contents:

1. INTRODUCTION
1.1. EEG-based brain-computer interfaces
1.2. The EEG feature space
1.3. Blind source separation of EEG
1.4. Classification and the feature selection problem
2. METHODS
2.1. Subjects
2.2. BCI experiment procedure
2.3. EEG recording and feature composition
2.4. Classification
2.5. Feature selection
3. RESULTS
3.1. Fitness evolution and overfitting at the feature selection level
3.2. The benefit of feature selection
3.3. The benefit of BSS transformations
3.4. Intersubject variability in good feature subsets
3.5. Feature values corresponding to the “yes” and “no” trials
4. DISCUSSION
4.1. Feature selection in the EEG-based BCI
4.2. The classifier and subset search parameter space
4.3. BSS in EEG-based BCI
4.3.1. “Good” feature subsets
4.4. BCI application relevance
5. CONCLUSION
ACKNOWLEDGMENTS
REFERENCES