EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

SincEEGNet: An end-to-end adaptive multi-frequency convolutional neural network for SSVEP recognition

Dai Wang, Le Wu, Bo Xue, Aiping Liu

Intelligent Sports and Health · 2025

DOI: 10.1016/j.ish.2025.05.002 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Intelligent Sports and Health, 2025. Verfasst von Dai Wang und 3 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recently, the filter banks (FB) strategy was utilized in the design of deep learning (DL) methods to recognize the steady-state visual evoked potentials (SSVEP). The methods with FB can leverage the multi-frequency rhythms contained in the SSVEP to achieve better recognition performance. However, the existing methods separately implemented FB in the signal preprocessing stage and the FB parameters were pre-defined and fixed, which undermined the end-to-end characteristics of the DL methods. Meanwhile, the design of pre-defined FB heavily relied on experts experience and the designed FB may be not optimal. These drawbacks restricted the existing methods from fully utilizing the SSVEP rhythms. In this study, we adopted the SincNet module to design an end-to-end adaptive multi-frequency neural network (NN) named SincEEGNet for SSVEP recognition. The adaptive band-pass FB realized by SincNet were integrated into the SincEEGNet to divide the input SSVEP into multi-frequency sub-bands. Then the features of each sub-band were extracted by NN and mapped to the categorical probability. In the design of SincEEGNet, the parameters of FB were learned by the back-propagation. It allowed the SincEEGNet to adaptively learn and build the optimal band-pass filters according to the input data during training. Thus, the SincEEGNet can achieve better utilization of the multi-frequency rhythms contained in the input SSVEP. In this study, we verified that SincEEGNet significantly outperformed the benchmark methods. In addition, we analyzed the effectiveness and rationality of the design of the SincEEGNet model through filter visualization and ablation experiments. Our study will contribute to the further exploration of the SSVEP recognition methods. • Based on the SincNet, an adaptive multi-frequency module named SincEEGNet was proposed for SSVEP recognition. • The effectiveness of the adaptive filter-banks of SincEEGNet was validated through the visualization. • The design methods of filter-banks for SSVEP recognition were discussed.

Zitieren

APA

Wang, D., Wu, L., Xue, B., & Liu, A. (2025). SincEEGNet: An end-to-end adaptive multi-frequency convolutional neural network for SSVEP recognition. Intelligent Sports and Health. https://doi.org/10.1016/j.ish.2025.05.002

Vancouver

Wang D, Wu L, Xue B, Liu A. SincEEGNet: An end-to-end adaptive multi-frequency convolutional neural network for SSVEP recognition. Intelligent Sports and Health. 2025. doi:10.1016/j.ish.2025.05.002

BibTeX
@article{Wang2025,
  author = {Wang, Dai and Wu, Le and Xue, Bo and Liu, Aiping},
  title = {SincEEGNet: An end-to-end adaptive multi-frequency convolutional neural network for SSVEP recognition},
  journal = {Intelligent Sports and Health},
  year = {2025},
  doi = {10.1016/j.ish.2025.05.002}
}
Kostenlos starten →