Transfer Adaptive Dictionary Learning With Intraclass Low-Rank Regularization for EEG Signal Classification

Classification of electroencephalogram (EEG) signals holds significant implications for assisting clinical diagnosis, treatment, and monitoring. Nonetheless, this domain encounters several challenges arising from the diversity, complexity, and paucity of EEG data. Therefore, this article proposes transfer adaptive dictionary learning with intraclass low-rank regularization (TADL-ICLR) for EEG signal classification. Within the framework of multisource domain transfer dictionary learning, TADL-ICLR employs projection matrices to project multiple source domains and target domain into a common subspace. This algorithm seeks a common dictionary across different domains to extract underlying data information, allowing data from various domains to be represented by similar sparse coding. Based on multidomain projected data and their sparse coding, TADL-ICLR first establishes an adaptive local linear embedding term to uncover the intrinsic geometric structure of data across domains. Second, TADL-ICLR introduces intraclass low-rank regularization, which imposes a low-rank structure on sparse coding with class information to counteract the blindness of the common dictionary and uncover latent class-discriminative information in the subspace. Third, TADL-ICLR incorporates an adaptive classifier with active samples, utilizing not only labeled samples but also the unlabeled samples in the target domain to enhance the dictionary’s discriminative power. Experimental results on public datasets demonstrate that the proposed algorithm outperforms other state-of-the-art algorithms.