基于通道动态优化与特征重用的多尺度DenseNet脑电情绪识别

打开文本图片集

[中图分类号]TP391.4;TP 183[文献标志码]A[文章编号]1005-0310（2026）01-0041-08

Abstract：To addressthe insufficient reuse of shallow features and static channel modeling in existing EEG-based emotion recognition models，this study proposes a channel-optimized DenseNet framework．Byincorporating squeeze-and-excitation（SE）modules to dynamically adjust weights of key prefrontal-parietal chanels and leveraging multi-scale convolutional kernels （ （1×1.3×3.5×5） for fusing δ/θ band differential entropy features， the model significantly enhances shallow feature utilization while suppressing noise interference.In the single subject experiments on the SEED dataset， the model achieved an accuracy of 96.73% ， significantly outperforming the baseline models （DBN： 86.08% ; DGCNN ： 90.40% ），and demonstrated robust performance across different channel configurations.

Keywords ： electroencephalogram （EEG） signals； channel adaptation; feature reuse; squeeze-and-excitation（SE）module;dynamic weights

0 引言

情绪反映了人类对客观事物的态度与行为响应，它在心理健康监测[1]、人机交互[2]等领域具有重要应用价值。（剩余11050字）