Deep learning has achieved a remarkable success in areas such as brain-computer interface systems (BCI). However, electroencephalography (EEG) signals evoked by motor imagery (MI) are sometimes limited in their amount due to invalid data caused by the subjects’ fatigue, leading to a performance degradation. To this end, in this work we extend empirical mode decomposition into multivariate empirical mode decomposition and intrinsic time-scale decomposition, proposing a component-mixing strategy (CMS) for MI data augmentation. Compared to commonly used data augmentation methods such as generative adversarial networks, CMS can generate artificial trials from a few training samples without any required training. We claim that raw and artificial data generated by CMS are consistent with respect to the distribution and power spectral density. Experiments done on the BCI Competition IV dataset 2b show that CMS can achieve a considerable improvement on the binary classification accuracy and the area under the curve score using EEGNet, wavelet neural networks and a support vector machine.

深度学习在脑机接口系统（BCI）等领域取得了显着的成功。然而，由于受试者疲劳引起的无效数据，运动意象（MI）诱发的脑电图（EEG）信号的数量有时会受到限制，从而导致性能下降。为此，在这项工作中，我们将经验模式分解扩展为多元经验模式分解和内在时间尺度分解，提出了一种用于 MI 数据增强的组件混合策略（CMS）。与常用的数据增强方法（如生成对抗网络）相比，CMS 可以从少量训练样本中生成人工试验，而无需任何训练. 我们声称 CMS 生成的原始数据和人工数据在分布和功率谱密度方面是一致的。在 BCI Competition IV 数据集 2b 上进行的实验表明，CMS 可以使用 EEGNet、小波神经网络和支持向量机在二元分类精度和曲线下面积得分上取得相当大的提高。