Scholarlink

<p>The practical deployment of real-time brain–computer interface (BCI) systems is constrained by the significant computational cost of accurately decoding motor imagery (MI) on portable, resource-limited hardware. This inefficiency stems from conventional CNNs failing to address the inherent feature redundancy in electroencephalography (EEG) signals. We demonstrate the effectiveness of AgileEEG by using it to enable real-time, closed-loop control of a portable upper-limb rehabilitation exoskeleton powered by a Raspberry Pi 4B. On a public dataset, AgileEEG achieved an accuracy of 50.25% on a six-class motor imagery task, a significant 6.72% improvement over EEGNet with substantially lower computational cost. In real-time, closed-loop online tests using a task decomposition strategy for four functional movements, our system reached a functional accuracy of 70.56%. AgileEEG offers a practical and computationally efficient solution, enabling advanced BCI control for assistive devices like rehabilitation exoskeletons.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AgileEEG: a lightweight CNN enabling real-time BCI control of a portable rehabilitation exoskeleton

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Yifan Wang,
  • Jiangfan Yao

摘要

The practical deployment of real-time brain–computer interface (BCI) systems is constrained by the significant computational cost of accurately decoding motor imagery (MI) on portable, resource-limited hardware. This inefficiency stems from conventional CNNs failing to address the inherent feature redundancy in electroencephalography (EEG) signals. We demonstrate the effectiveness of AgileEEG by using it to enable real-time, closed-loop control of a portable upper-limb rehabilitation exoskeleton powered by a Raspberry Pi 4B. On a public dataset, AgileEEG achieved an accuracy of 50.25% on a six-class motor imagery task, a significant 6.72% improvement over EEGNet with substantially lower computational cost. In real-time, closed-loop online tests using a task decomposition strategy for four functional movements, our system reached a functional accuracy of 70.56%. AgileEEG offers a practical and computationally efficient solution, enabling advanced BCI control for assistive devices like rehabilitation exoskeletons.