Attention-Based Parallel Multiscale Convolutional Neural Network for Visual Evoked Potentials EEG Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) decoding is crucial for Visual Evoked Potentials-based Brain-Computer Interfaces (BCIs).
• User fatigue from repetitive stimuli degrades EEG signal quality and BCI performance.
• Existing methods struggle with reduced attention and increased noise in fatigued states.

Purpose of the Study:

• To develop a novel deep learning model to enhance EEG decoding accuracy in BCIs.
• To address the challenges posed by user fatigue on BCI performance.
• To improve the robustness and reliability of Visual Evoked Potentials decoding.

Main Methods:

• Proposed an attention-based parallel multiscale convolutional neural network (AMS-CNN).
• Utilized parallel convolutional layers with varying temporal filters for robust feature extraction.
• Employed sequential convolution blocks for spatial and temporal fusion, incorporating an attention mechanism for feature weighting.

Main Results:

• The AMS-CNN demonstrated superior classification performance on fatigue datasets compared to state-of-the-art methods.
• Analysis confirmed the effectiveness of multiscale convolution and attention mechanisms in improving decoding.
• The proposed framework significantly enhanced EEG decoding in fatigued subjects.

Conclusions:

• The AMS-CNN offers a promising solution for improving Visual Evoked Potential BCI decoding performance.
• The model effectively mitigates the negative impact of user fatigue on EEG signal reliability.
• This approach has the potential to enhance user experience and broaden BCI applications.