Lightweight neural network design for real-time signal processing in brain-computer interfaces
Caiyun Shan
Communication University of China, Nanjing
Yanbin Bu
Communication University of China, Nanjing
DOI: https://doi.org/10.59429/esta.v12i2.10566
Keywords: Brain-computer interface; Real-time signal processing; Lightweight neural network; EEGNet
Abstract
Traditional signal processing methods struggle to meet the requirements of real-time performance and accuracy, while deep learning-based approaches, despite their superior performance, suffer from high computational complexity, which hinders real-time processing on embedded devices. To address this challenge, this paper explores the design of lightweight neural networks for real-time signal processing in brain-computer interfaces (BCIs). By analyzing the characteristics of electroencephalogram (EEG) signals, we summarize key considerations for lightweight network design and present an efficient EEG signal classification model using EEGNet as an example. Finally, we discuss the prospects and challenges of lightweight neural networks in BCIs, providing insights for future research in this field.
References
[1]He, Q., Wu, H., Tian, F., et al. A real-time brain-computer interface system for controlling short message transmission. Chinese Journal of Medical Physics, 2012, 29(3): 3386-3392.
[2]Zhu, X., Han, Y., Hao, Y., et al. Real-time spike sorting algorithm based on probabilistic neural network. Journal of South China University of Technology (Natural Science Edition), 2012, 40(6): 48-55.
[3]Xie, Y., Liu, Y., Chen, C., et al. An embedded acceleration method for lightweight convolutional neural networks. Journal of Chinese Computer Systems, 2023, 44(7): 1345-1351.
[4]Zhang, T., Xu, D., Chen, J., et al. Patent main path analysis in brain-computer interface field. China Medical Devices, 2025, 40(3): 8-15.