Research on data fusion group-brain motor imagery technology based on CNN+BiLSTM model with DJ

Abstract:

To Address the challenges of large data volumes, individual data variability, and asynchronous EEG signals in group-brain Motor Imagery Brain-Computer Interface (MI-BCI) decoding tasks, this study proposes serial and parallel data fusion methods suitable for group-brain MI-BCI, and CNN+BiLSTM with DJ decoding model. In this model, CNN can adapt to the extraction of spatiotemporal features from large datasets of group-brain BCI multi-person data, while BiLSTM (Bi-directional LSTM) further maintains the long-term temporal dependencies of EEG data. The Data Jitter (DJ) helps exactly pinpoint MI data, improving the classification accuracy and robustness of the model. This paper utilizes public datasets for training and validating both serial and parallel models for group-brain data fusion and designs an ex-perimental paradigm to collect a dataset for testing model performance. Results show that compared to single-person sce-narios, the serial fusion group-brain model exhibits higher accuracy, whereas the accuracy of the parallel data model is lower than that of the single person model.

Key words:Group-Brain BCI, Motor Imagery, Data Fusion, Data Jitter, Deep Learning