Scholarlink

<p>Scalp electroencephalography (scalp-EEG) reliably extracts steady-state visual evoked potentials (SSVEP) by monitoring occipital cortex activity. However, ear-EEG offers a more convenient, wearable alternative, yet its effectiveness for SSVEP detection requires further validation. In this study, we propose a lightweight ensemble deep learning architecture for SSVEP classification using ear-EEG signals. The model incorporates multiple temporal windows to capture diverse pattern representations, reducing individual model errors. We evaluated our method on a public ear-EEG dataset collected from eleven subjects across three separate sessions. The proposed model substantially outperformed canonical correlation analysis (CCA), achieving average accuracies of 79.57% versus 34.79% for session 2, and 81.32% versus 34.39% for session 3. Compared to the state-of-the-art E-EEGNet model, our approach achieved comparable classification accuracy while requiring fewer than 50% of the trainable parameters (61,711 versus 133,755). Individual subject accuracies reached as high as 97.4% and 96.59% in sessions 2 and 3, respectively. These results demonstrate that computationally efficient deep learning can achieve accurate SSVEP classification from ear-EEG, supporting the development of practical, wearable ear-EEG-based brain-computer interfaces.</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

Ear-EEG for SSVEP classification: a computationally efficient deep learning approach

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Mahdi Abbasi,
  • Mohammad Bagher Khodabakhshi,
  • Shahriar Jamasb

摘要

Scalp electroencephalography (scalp-EEG) reliably extracts steady-state visual evoked potentials (SSVEP) by monitoring occipital cortex activity. However, ear-EEG offers a more convenient, wearable alternative, yet its effectiveness for SSVEP detection requires further validation. In this study, we propose a lightweight ensemble deep learning architecture for SSVEP classification using ear-EEG signals. The model incorporates multiple temporal windows to capture diverse pattern representations, reducing individual model errors. We evaluated our method on a public ear-EEG dataset collected from eleven subjects across three separate sessions. The proposed model substantially outperformed canonical correlation analysis (CCA), achieving average accuracies of 79.57% versus 34.79% for session 2, and 81.32% versus 34.39% for session 3. Compared to the state-of-the-art E-EEGNet model, our approach achieved comparable classification accuracy while requiring fewer than 50% of the trainable parameters (61,711 versus 133,755). Individual subject accuracies reached as high as 97.4% and 96.59% in sessions 2 and 3, respectively. These results demonstrate that computationally efficient deep learning can achieve accurate SSVEP classification from ear-EEG, supporting the development of practical, wearable ear-EEG-based brain-computer interfaces.