HCFNet: A heterogeneous frequency bands coupling CNN for enhanced short-time fast response in motor imagery decoding

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery signals contain diverse frequency components.
• Current methods often process frequency bands uniformly, losing crucial information and introducing redundancy.
• Heterogeneity and coupling of frequency bands are overlooked in existing models.

Purpose of the Study:

• To develop a novel method for motor imagery signal decoding that addresses limitations of existing approaches.
• To explore heterogeneous feature extraction and coupling across frequency bands for improved precision.
• To enhance the performance of brain-computer interfaces (BCIs) through advanced signal processing.

Main Methods:

• Proposed HCFNet model for heterogeneous feature extraction and cross-frequency coupling.
• Separated raw signals into high and low-frequency bands with specialized modules.
• Employed a cross-frequency coupling module and data augmentation for robust spectral-spatiotemporal feature capture.

Main Results:

• Achieved average accuracies of 82.41% on BCIC-IV-2a and 76.52% on OpenBMI.
• HCFNet demonstrated superior performance compared to state-of-the-art methods.
• Maintained excellent performance even with shorter time windows, indicating efficiency.

Conclusions:

• HCFNet significantly advances motor imagery signal decoding technology.
• The innovative frequency-band heterogeneous coupling method offers substantial potential for rapid responses in BCIs.
• Future applications include domain adaptation, cross-domain alignment, and cross-subject contexts.