Epileptic eeg data augmentation method based on discrete state space embedding diffusion model
By using a discrete state space embedding diffusion model and combining VAE and VQ-VAE dual encoding mechanisms, diverse epilepsy EEG samples are generated, solving the problems of data class imbalance and generation instability, and significantly improving the accuracy of epilepsy detection.
CN122153438APending Publication Date: 2026-06-05TAIYUAN UNIVERSITY OF TECHNOLOGY
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TAIYUAN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-05
Smart Images
Figure CN122153438A_ABST
Abstract
The application discloses a kind of epilepsy electroencephalogram data enhancement methods based on discrete state space embedding diffusion model, comprising the following steps: S1, constructs epilepsy electroencephalogram time-frequency feature dataset;S2, constructs based on discrete state space embedding diffusion model: based on VAE encoder architecture, constructs continuous latent space representation extraction module;Based on VQ-VAE encoder architecture, constructs discrete latent space representation extraction module;Based on U-Net architecture, diffusion model backbone network architecture is constructed, and generation time-frequency feature is obtained;S3, design dynamic denoising enhancement strategy, change diffusion denoising step number is to each original sample generates several enhanced variants;S4, input the mixed dataset after enhancement into deep neural network training, select the optimal model as the final model weight, and compare the results with traditional data enhancement method.The application improves the sensitivity and specificity of epilepsy detection model under the condition of severe data imbalance.
Need to check novelty before this filing date? Find Prior Art