Scholarlink

In this study, a multi-scale feature fusion CNN-LSTM network (MFFCL) is proposed, aiming at recognizing multi-class EEG signals from motion images. The network divides the frequency bands of EEG signals by overlapping filter sets and combines the cascaded CNN-LSTM structure to effectively fuse the spatial, temporal and frequency features of EEG signals. It is shown that MFFCL can effectively separate the discriminative features in multi-frequency components and accurately identify the spatial and temporal features of EEG data, which is highly adaptable in processing small-scale EEG data. Preliminary results show that the average accuracy of MFFCL on multiple datasets is as high as 88% or more, which proves its effectiveness and utility in the field of EEG signal decoding, and provides a powerful tool for feature extraction and classification in MI-BCI research.

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

Multi-scale CNN-LSTM Network for Motor Imagery EEG Signals Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xinhang Gu,
  • Hui Peng,
  • Xuedou Xiong,
  • Chaoyue Wang,
  • Juan Yang

摘要

In this study, a multi-scale feature fusion CNN-LSTM network (MFFCL) is proposed, aiming at recognizing multi-class EEG signals from motion images. The network divides the frequency bands of EEG signals by overlapping filter sets and combines the cascaded CNN-LSTM structure to effectively fuse the spatial, temporal and frequency features of EEG signals. It is shown that MFFCL can effectively separate the discriminative features in multi-frequency components and accurately identify the spatial and temporal features of EEG data, which is highly adaptable in processing small-scale EEG data. Preliminary results show that the average accuracy of MFFCL on multiple datasets is as high as 88% or more, which proves its effectiveness and utility in the field of EEG signal decoding, and provides a powerful tool for feature extraction and classification in MI-BCI research.