黄伟坤 谢伟
（广东工业大学，广东 广州 510006）
摘要：针对基于发音想象的脑机接口样本数据数量小、数据噪声大，导致模型泛化能力差的问题，提出一种基于跨模态信息迁移的发音想象脑电信号分类方法。该方法通过知识蒸馏，将音频模态信息迁移到脑电模态，从而提高模型的泛化能力；通过多尺度学习来提高模型性能。在数据集Kara One中，两个二分类任务的AUC分别为68.28%和69.53%。实验结果表明，该方法有效地提高了模型的性能。
关键词：发音想象；脑机接口；跨模态；知识蒸馏；信息迁移
中图分类号：TP391 文献标志码：A 文章编号：1674-2605(2023)03-0004-06
DOI：10.3969/j.issn.1674-2605.2023.03.004
Classification Method of EEG Signals of Pronunciation in Imagined Based on Cross Modal Information Transfer
HUANG Weikun XIE Wei
(Guangdong University of Technology, Guangzhou 510006, China)
Abstract: Aiming at the problem that the datasets of brain-computer interface based on pronunciation in imagined is small and the data noise is loud, which leads to the poor generalization ability of the model, a classification method of EEG signals of pronunciation in imagined based on cross modal information transfer is proposed. In this method, the audio modal information is transferred to the EEG modality by knowledge distillation, so as to improve the generalization ability of the model. The method also improves the performance of the model through multi-scale learning. In the dataset Kara One, the AUC of two binary classification tasks is 68.28% and 69.53%, respectively. Experimental results demonstrate that this method effectively enhances the performance of the model.
Keywords: pronunciation in imagined; brain-computer interface; cross modal; knowledge distillation; information transfer
