EEG Channel Localization and Selection via Training with Noise Injection for BCI Applications

Electroencephalography (EEG) is crucial for monitoring brain activity in neuroscience and clinical applications. However, the multitude of channels recorded by scalp electrodes poses challenges, including impractical usage and high model complexity. This paper addresses the challenges of high dimensionality in EEG data and introduces an innovative EEG channel selection algorithm, LSvT-NI, based on model training and noise injection, achieving substantial reductions in channels, model size, and complexity while maintaining high classification accuracy. Validated through experiments on EEGNet and the BCI Competition IV 2a dataset, the algorithm proves beneficial for practical and cost-efficient scenarios. Specifically, experiments on the BCI Competition IV 2a dataset demonstrate that LSvT-NI with white noise and pink noise at 5dB SNR achieves a remarkable 77.3% and 72.7% reduction in channels, along with 11.7% and 11% reductions in model size, and 86.9% and 71.8% in computation complexity.