Abstract:
Multi-channel electroencephalography (EEG), as a non-invasive technique, records brain electrical activity through multiple electrodes placed on the scalp, aiding in understanding individual psychological states and assisting in the diagnosis of various diseases. Given the high cost and technical challenges of annotating large-scale EEG data, self-supervised learning (SSL), as a label-free learning paradigm, has garnered widespread attention in the EEG domain. SSL leverages the intrinsic structure of data for learning, effectively enhancing model generalization ability. Despite significant advancements in current research, it encounters the following challenges: firstly, multi-channel EEG data contains complex spatiotemporal correlations, yet many existing methods are confined to modeling either temporal or spatial dimensions independently, rather than integrating both comprehensively and thus limiting a profound understanding of the intrinsic complexities of EEG signals; secondly, many current approaches have failed to integrate segment-level and instance-level information effectively. The former contributes to improving the model's generalization capability, while the latter aids the model in better adapting to downstream (classification) tasks. To address the aforementioned challenges, this study proposes and implements a self-supervised pre-training framework that combines contrastive and reconstruction strategies. Specifically, by introducing a channel masking strategy on the basis of temporal dimension mask reconstruction, the spatiotemporal relationships of EEG data are effectively captured. Through fine-grained capturing of inter-segment relationships, the performance and generalization ability of the model are improved. Meanwhile, combining instance-level contrastive learning with masked reconstruction tasks helps the model learn representations with instance discriminability and local perception ability, thus better adapting to downstream tasks. Furthermore, the introduction of a self-paced learning mechanism further enhances the stability of model training. Finally, experimental results across multiple EEG tasks validate the effectiveness of the proposed approach.