Computing and Informatics

Deep Learning-based electroencephalogram (EEG) decoding in brain-computer interface (BCI) faces many challenges: First, EEG has low spatial resolution, which makes it difficult to accurately identify specific brain regions; Second, deep learning requires large amounts of data, while EEG datasets are typically small. Merging small datasets raises privacy concerns because EEG signals contain a wealth of personal information. Finally, while federated learning (FL) offers a potential solution for addressing the issues of limited data and privacy, it faces the challenge of non-independent and identically distributed (non-IID) data, as the local data on the client is often derived from datasets of different tasks or subjects. To solve these problems, this paper proposes a novel Federated Graph Learning (FGL) framework based on Electroencephalogram and Electromyography (EEG-EMG), named FSDFGL. During the construction of the graph data, EMG data is incorporated into EEG data through SPMI, as EMG signals provide distinct features of muscle activity. In the local Graph Neural Network (GNN) process, the structural information is decoupled from the node features. Then, the structure encoder is used to capture the structural information and share it globally during FL. This allows FSDFGL to capture more commonly shared structural information while avoiding degradation of local model performance due to sharing heterogeneous features. The framework was evaluated on the TAN dataset, and experimental results demonstrate the superiority of FSDFGL in identifying complex motion intentions.