| 毛传波,杨庆华,蔡世波,王志恒.基于独立分量分析和普鲁克分析的运动想象迁移学习策略[J].高技术通讯(中文),2023,33(7):683~691 |
| 基于独立分量分析和普鲁克分析的运动想象迁移学习策略 |
| Transfer learning strategy for motor imagery based on independent component analysis and Procrustes analysis |
| |
| DOI：10. 3772/ j. issn. 1002-0470. 2023. 07. 002 |
| 中文关键词: 运动想象； 脑机接口； 独立分量分析（ICA）； 普鲁克分析； 迁移学习 |
| 英文关键词: motor imagery, brain-computer interface, independent component analysis (ICA), Procrustes analysis, transfer learning |
| 基金项目: |
| 作者 | 单位 | | 毛传波 | (浙江工业大学特种装备制造与先进加工技术教育部/浙江省重点实验室杭州 310023) | | 杨庆华 | | | 蔡世波 | | | 王志恒 | |
|
| 摘要点击次数: 724 |
| 全文下载次数: 635 |
| 中文摘要: |
| 针对传统上基于运动想象的脑机接口系统在应用之前需要进行枯燥冗长的校准实验的问题，提出基于独立分量分析（ICA）和普鲁克分析的迁移学习算法。该算法采用独立分量分析对脑电信号进行空间滤波，对比应用多种对齐变换方法，实现对样本数据的平移和缩放。并提出欧氏空间下的旋转变换方法，以进一步匹配目标受试者与其他受试者的样本分布，实现有效的跨受试者和跨数据集的迁移学习。所提算法相较基于黎曼普鲁克分析的方法具有更好的分类性能以及计算效率，并在公共数据集PhysionetMI和BCI-IV-2a上较传统机器学习算法将平均kappa值提高了约0.1和0.04。结果表明该方法能有效提高小样本下的分类正确率，这有助于降低对校准数据量的需求，从而减少校准实验的耗时。 |
| 英文摘要: |
| Aiming at the problem that the traditional brain-computer interface system based on motor imagery needs to perform a boring and lengthy calibration session before being used, a transfer learning algorithm based on independent component analysis (ICA) and Procrustes analysis is proposed. First, the electroencephalogram (EEG) signal is spatially filtered by independent component analysis, then a variety of alignment methods are used and compared with translate and scale the samples. And a rotation transformation method in Euclidean space is proposed to further match the data distribution of the target subject and other subjects, which helped to achieve effective transfer learning cross-subject and cross-dataset. The proposed method has better classification performance and computational efficiency than the method based on Riemannian Procrustes analysis, increasing the average kappa by 0.1 and 0.04 on the public dataset PhysionetMI and BCI-IV-2a compared with traditional machine learning algorithms. The results show that the method can effectively improve the classification accuracy under a small number of samples, which helps to reduce the demand for calibration data, thereby reducing the time required for the calibration session. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
