Scholarlink

In speech imagery brain-computer interface (BCI) research, two key limitations remain: most studies rely solely on standard convolutional neural networks, and few incorporate attention mechanisms. To address these gaps, we propose CSFA-PMCCNN, a novel deep learning model for EEG-based speech imagery decoding. The model integrates parallel multi-scale convolutional layers, a capsule network module, and a channel-space-frequency attention mechanism to enhance feature representation. Experimental results on the KaraOne dataset demonstrate that CSFA-PMCCNN improves average accuracy by 1.26% over the baseline CapsK-SI model. Furthermore, it achieves performance comparable to, and in some cases better than, several state-of-the-art models, confirming its effectiveness and generalization potential.

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

CSFA-PMCCNN: A Parallel Multiscale Convolutional Capsule Neural Network Based on Channel-Space-Frequency Attention Mechanism for Speech Imagery EEG Signals Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ke Su,
  • Liang Tian

摘要

In speech imagery brain-computer interface (BCI) research, two key limitations remain: most studies rely solely on standard convolutional neural networks, and few incorporate attention mechanisms. To address these gaps, we propose CSFA-PMCCNN, a novel deep learning model for EEG-based speech imagery decoding. The model integrates parallel multi-scale convolutional layers, a capsule network module, and a channel-space-frequency attention mechanism to enhance feature representation. Experimental results on the KaraOne dataset demonstrate that CSFA-PMCCNN improves average accuracy by 1.26% over the baseline CapsK-SI model. Furthermore, it achieves performance comparable to, and in some cases better than, several state-of-the-art models, confirming its effectiveness and generalization potential.