一种基于稀疏动态图卷积的脑电信号解码方法及系统
By employing a sparse dynamic graph convolution method, an adaptive adjacency matrix is constructed using the covariance matrix and bilinear mapping. Combined with an autoregressive moving average filter and fuzzy label learning, this approach addresses the issues of insufficient generalization ability and overfitting in existing EEG signal decoding models, achieving efficient classification of EEG signals and accurate modeling of individual-specific functional connections.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2025-05-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively characterize the complex spatial topological relationships between multi-channel EEG signals, and static adjacency matrices lack the ability to adaptively learn individual-specific neural response patterns, leading to insufficient model generalization ability and the risk of overfitting.
We employ a sparse dynamic graph convolution method to construct an adaptive adjacency matrix through covariance matrix and bilinear mapping. Combined with autoregressive moving average filter and fuzzy label learning, we dynamically generate a task-driven adjacency matrix and introduce sparsity constraints to improve the model's ability to focus on key brain region interactions and its robustness.
It improves the classification accuracy and generalization performance of EEG signal decoding models, enhances the ability to capture individual-specific functional connectivity patterns, and reduces noise interference and overfitting risks.
Smart Images
Figure CN120579029B_ABST
Abstract
Citation Information
Patent Citations
Brain-computer interface system based on Riemannian manifold discrimination
CN116804895A
Self-adaptive subgraph generation model construction method for EEG channel screening
CN117653139A