| 摘 要: 为了解决抑郁症检测中由于脑电(Electroencephalogram,EEG)信号数据稀缺导致的分类精度不高的问题,结合去噪扩散概率模型(Denoising Diffusion Probabilistic Model,DDPM)和卷积注意力机制(Convolutional
Block Attention Module,CBAM),提出了一种用于EEG数据增强的卷积注意力扩散模型(Convolutional Attention Diffusion Model,CADM)。在MODMA数据集上,生成了与原始EEG信号数据分布相似的时频特征,以扩充训练
数据集,并利用 Vision Transformer、MLP和SVM模型进行抑郁症分类。实验结果显示,增强后的数据集在3个分类器上的准确率分别提高了10.99%、3.88%和5.50%,证明模型生成的特征具有良好的鲁棒性,有效提升了EEG抑
郁症检测的准确性和稳定性。 |
| 关键词: 脑电信号 抑郁症检测 去噪扩散概率模型 数据增强 卷积注意力模块 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(61806118);陕西科技大学博士科研启动基金项目(2020BJ-30) |
|
| Convolutional Attention Diffusion Model for EEG Data Augmentation in Depression Detection |
|
WANG Zikai, CHEN Jingxia, JIA Xiaowen, ZHAO Chen
|
(School of Electronic Information and Artificial Intelligence, Shaanxi University of Science & Technology, Xi’an 710021, China)
221612066@sust.edu.cn; chenjingxia@sust.edu.cn; 221612107@sust.edu.cn; 221612091@sust.edu.cn
|
| Abstract: To address the issue of low classification accuracy caused by scarce EEG data in depression detection,this paper proposes a Convolutional Attention Diffusion Model (CADM) for EEG data augmentation by integrating
Denoising Diffusion Probabilistic Models (DDPM) and Convolutional Block Attention Module (CBAM). Using the MODMA dataset, the model generates time-frequency features consistent with the distribution of original EEG data to
expand the training dataset. Classification was then performed using Vision Transformer, MLP, and SVM models.Experimental results demonstrate that the augmented dataset improved classification accuracy by 10.99% , 3.88% and
5.50% across the three classifiers respectively. This proves the model’s robustness in feature generation and its effectiveness in enhancing the accuracy and stability of EEG-based depression detection. |
| Keywords: EEG signals depression detection denoising diffusion probabilistic models ( DDPM) data augmentation convolutional block attention module (CBAM) |
