A Lightweight Spiking Neural Network Model for Real-Time Brain Signal Classification Using Open EEG Datasets

Abstract

To classify EEG signals in real time, a lightweight SNN was built and evaluated. The work showed that it is possible to use energy-efficient, bio-inspired neural computer models on BCI devices using open-source EEG data. The preliminary results indicate that the proposed system's accuracy and speed are promising for implementation on a portable, low-power device. Due to their event-based computing paradigm and temporal coding feature, spiking neural networks (SNNs) have been gaining popularity in brain signal processing. A biologically plausible and efficient implementation of an SNN model was presented for the classification of EEG signals with an application to motor imagery tasks. The model proposed utilized the hybrid coding and attention mechanism to extract the spatiotemporal features in the EEG data and select the relevant features. High classification accuracy, low inference latency, and satisfactory cross-subject generalization performance were achieved by the model in large-scale experiments using publicly available EEG datasets. The results achieved validate the potential of SNNs as a promising alternative to conventional NNs for BCI applications. This result is a significant advancement in low-power, real-time neural decoding systems and opens the door for future generations of neuromorphic computing applications in the biomedical domain.