[1]孟颍辉,宋娇帅,李端,等.结合 SCHO 通道选择和 Top-k 注意力的 EEG 情绪识别[J].郑州大学学报(工学版),2027,48(XX):1-9.[doi:10. 13705 / j. issn. 1671-6833. 2027. 02. 002]
　MENG Yinghui,SONG Jiaoshuai,LI Duan,et al.EEG Emotion Recognition Combining SCHO Channel Selection and Top-k Attention[J].Journal of Zhengzhou University (Engineering Science),2027,48(XX):1-9.[doi:10. 13705 / j. issn. 1671-6833. 2027. 02. 002]
点击复制

结合 SCHO 通道选择和 Top-k 注意力的 EEG 情绪识别()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
48
期数:
2027年XX
页码:
1-9
栏目:
出版日期:
2027-12-10

文章信息/Info

Title:
EEG Emotion Recognition Combining SCHO Channel Selection and Top-k Attention
作者:
孟颍辉, 宋娇帅, 李端, 南姣芬, 冯文, 朱付保, 苑常香
郑州轻工业大学 计算机与人工智能学院,河南 郑州 450001
Author(s):
MENG Yinghui , SONG Jiaoshuai , LI Duan , NAN Jiaofen , FENG Wen , ZHU Fubao , YUAN Changxiang
School of Computer Science and Artificial Intelligence, Zhengzhou University of Light Industry, Zhengzhou 450001,China
关键词:
脑电信号;  情绪识别;  残差网络;  多头注意力
Keywords:
EEG signals;  emotion recognition;  residual network;  multi-head attention
分类号:
TN911. 7 R318
DOI:
10. 13705 / j. issn. 1671-6833. 2027. 02. 002
文献标志码:
A
摘要:
情绪识别是脑机接口、情感计算和心理健康监测中的重要研究内容,现有 EEG 情绪识别系统普遍存在电极数量较多、采集成本较高、可穿戴性不足以及低通道条件下识别性能下降等问题。 为解决上述问题,本文提出了一种结合元启发式优化算法 SCHO 与 CRTA 框架的低通道 EEG 情绪识别算法,实现了自适应通道选择与高效特征建模的统一。 首先,对 EEG 信号进行快速傅里叶变换与功率谱密度特征提取,使 SCHO 能够针对不同受试者自适应筛选最具情绪判别力的电极子集。 随后,CRTA 通过卷积层与残差块逐层捕获多尺度特征,并在深层结构中引入改进的 Top-k 多头稀疏注意力机制,以抑制噪声冗余、突出关键情绪特征,从而提升识别效率与精度。 实验结果表明,所提出的算法在 SEED 与 DEAP 数据集上均取得了优异性能;即使仅使用 6 个参考电极( F3、F8、T7、T8、P3、P4) ,仍能保持较高的情绪识别准确率。 综上,本文算法为低通道、高性能 EEG 情绪识别提供了一种有效方案,并为轻量化、可穿戴 EEG 情绪监测系统设计提供了参考。
Abstract:
EEG-based emotion recognition was an important research topic in brain-computer interfaces, affective computing, and mental health monitoring. However, existing EEG emotion recognition systems generally suffered from a large number of electrodes, high acquisition costs, limited wearability, and performance degradation under low-channel conditions. To address these issues, a low-channel EEG emotion recognition algorithm was proposed, in which the meta-heuristic optimization algorithm Sinh Cosh Optimizer ( SCHO) was integrated with the convolutional residual Top-k attention framework ( CRTA) to achieve adaptive channel selection and effective feature modeling. First, fast Fourier transform and power spectral density features were extracted from EEG signals, and SCHO was employed to adaptively select subject-specific emotion-discriminative electrode subsets. Subsequently, multiscale features were captured through convolutional layers and residual blocks in CRTA, while an improved Top-kmulti-head sparse attention mechanism was introduced in deep layers to reduce redundant information and enhance emotion-related features. Experimental results on the SEED and DEAP datasets demonstrated that the proposed algorithm achieved competitive performance under both full-channel and low-channel settings. Furthermore, a fixed six-electrode configuration consisting of F3, F8, T7, T8, P3, and P4 was found to maintain high recognition accuracy, providing an effective solution for low-channel and high-performance EEG emotion recognition and offering a reference for lightweight wearable EEG emotionmonitoring systems.
更新日期/Last Update: 2026-10-30