A fused multi-subfrequency bands and CBAM SSVEP-BCI classification method based on convolutional neural network

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Traditional methods struggle with classifying short-time window steady-state visual evoked potential (SSVEP) signals for brain-computer interfaces (BCI).
• Accurate and efficient SSVEP signal processing is crucial for advancing BCI applications.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) based classification method, termed CBAM-CNN, for enhanced SSVEP-BCI tasks.
• To improve classification performance, particularly for short-duration SSVEP signals, by integrating multi-subfrequency bands and a convolutional block attention module (CBAM).

Main Methods:

• Extraction and fusion of multi-subfrequency bands SSVEP signals as initial network input.
• Integration of CBAM for adaptive feature refinement at both initial input and feature fusion stages.
• Validation using Inner Mongolia University of Technology (IMUT) and Tsinghua University (THU) datasets.

Main Results:

• The proposed CBAM-CNN achieved a maximum accuracy of 0.9813 percentage points (pp).
• CBAM-CNN demonstrated superior accuracy (0.0201-0.5388 pp higher) compared to CNN, CCA-CWT-SVM, CCA-SVM, CCA-GNB, FBCCA, and CCA within 0.1-2s time windows.
• Exceptional performance was observed in short time windows (0.1-1s), with a maximum information transmission rate (ITR) of 503.87 bit/min.

Conclusions:

• CBAM-CNN significantly outperforms existing methods in SSVEP decoding, especially under short time constraints.
• The method shows substantial improvements in accuracy and information transmission rate, highlighting its potential for practical SSVEP-BCI applications.