一种面向类别不平衡任务的脑电信号分类系统及方法
By constructing time-frequency embedding, bidirectional temporal Mamba, and time dynamic graph modules, the problem of class imbalance in EEG signal classification is solved, and multi-scale feature modeling and dynamic spatial dependence capture of EEG signals are achieved, improving classification accuracy and robustness, especially in the recognition capabilities of epileptic seizure type identification and sleep staging tasks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN UNIV
- Filing Date
- 2025-09-25
- Publication Date
- 2026-07-17
AI Technical Summary
Existing EEG signal classification methods suffer from several problems when facing class imbalance, including strong dependence on feature extraction, insufficient efficiency and long-range dependency modeling ability of deep learning methods in long sequence modeling, insufficient ability to model time-frequency features, and insufficient adaptation to class imbalance. These problems cause the model to be biased towards the majority class during training, making it difficult to effectively identify minority class features and affecting classification accuracy and robustness.
We employ a time-frequency embedding module for parallel time-domain and frequency-domain feature extraction, combined with a bidirectional temporal Mamba unit and a time dynamic graph construction module for EEG. Through a category-aware attention mechanism and an adaptive loss function, we enhance the representation ability of minority class samples, thereby achieving multi-scale feature modeling and dynamic spatial dependence capture of EEG signals.
It significantly improves the accuracy and robustness of EEG signal classification, especially in scenarios where minority class samples are scarce, such as epileptic seizure type identification and sleep stage segmentation, thereby enhancing the model's generalization ability and clinical applicability.
Smart Images
Figure CN121370187B_ABST