| 苏靖然,李秋生.基于VMD和融合通道注意力机制DenseNet的脑电情绪识别[J].井冈山大学自然版,2025,46(4):81-87 |
| 基于VMD和融合通道注意力机制DenseNet的脑电情绪识别 |
| EEG EMOTION RECOGNITION BASED ON VMD AND DENSENET WITH FUSION CHANNEL ATTENTION MECHANISM |
| 投稿时间：2025-02-11 修订日期：2025-03-07 |
| DOI：10.3969/j.issn.1674-8085.2025.04.010 |
| 中文关键词: EEG VMD 特征重用 多尺度卷积核 通道注意力机制 |
| 英文关键词: EEG VMD feature reuse multi-scale convolution kernel channel attention |
| 基金项目:国家自然科学基金项目(61561004)；江西省自然科学基金项目(20242BAB25052) |
|
| 摘要点击次数: 137 |
| 全文下载次数: 207 |
| 中文摘要: |
| 针对目前提取脑电信号特征比较困难以及其准确率受限的问题，提出一种新型的基于VMD分解和DenseNet网络融合通道注意力机制的脑电情绪识别模型，主要包括特征提取和分类两个模块。首先，引入VMD分解算法，一方面，可以实现对脑电信号的去噪处理，另一方面，在有效IMF分量上提取微分熵特征，提高不同情绪状态下特征的可判别性；其次，DenseNet网络中融入多尺度卷积核以及通道注意力机制，不仅能够提取不同尺度的特征，还可以根据不同脑电通道与情绪状态的关联程度对其赋予不同的权值，进一步提高情绪识别准确率；最后，在SEED数据集上验证该模型的有效性与鲁棒性，15个被试者的平均分类准确率可达96.86%。结果表明，提出新的模型能有效地提取与情绪相关的脑电特征，并可以取得较好的分类效果。 |
| 英文摘要: |
| Given the current difficulties of EEG signals feature extraction and limited accuracy, a new EEG emotion recognition model based on VMD and DenseNet with fusion channel attention mechanism is proposed, mainly including two modules: feature extraction and classification. Firstly, the VMD algorithm was introduced. It could realize the denoising processing of EEG signals, and the differential entropy features were extracted from the effective IMF signals to improve the discriminability of features under different emotional states. Secondly, the multi-scale convolution kernel and channel attention mechanism were integrated into the DenseNet network, which could not only extract the features of different scales, but also given different weights to different EEG channels, so as to further improve the accuracy of emotion recognition. Finally, the effectiveness and robustness of the model were verified on the SEED data set, and the average classification accuracy of 15 subjects could reach 96.86%. The results show that the proposed model can effectively extract EEG features related to emotion and achieve better classification results. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
