Multimodal Learning in Brain-Computer Interfaces: A Research Review on Applications

Brain-Computer Interfaces (BCIs) enable human-machine interaction by decoding neural activity, demonstrating broad application potential in fields such as medical rehabilitation and neural engineering.However, traditional unimodal BCI systems face limitations due to signal noise, individual variability, and task complexity, hindering their practical utility.Multimodal learning significantly enhances decoding accuracy, robustness, and generalization capabilities through the fusion of signals from multiple data sources, including electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), and eye-tracking.This paper analyzes research progress in this field, covering signal acquisition and preprocessing, fusion strategies at the data layer, feature layer, and decision layer, typical applications (such as emotion recognition, vigilance monitoring, and neural decoding), and cutting-edge directions (including zero-shot learning, federated learning, and generative augmentation).Research indicates that multimodal fusion can improve BCI system performance by 15%-25% and enhance stability in complex environments by over 30%.This review systematically summarizes the theoretical framework and technical pathways of multimodal BCI, offering key insights to facilitate its transition from laboratory research to clinical applications