基于GCN-LSTM融合模型的脑电情绪时空特征协同识别——融合频域特征的建模与识别性能分析

Based on GCN-LSTM Fusion Model for Synergistic Recognition of Spatiotemporal Features in EEG Emotion—Modeling with Fusion of Frequency-Domain Features and Recognition Performance Analysis

摘要: 本文围绕脑电情绪识别展开研究，针对脑电信号低信噪比、非平稳特性及现有模型在空间时序建模、跨被试泛化等方面的不足，提出融合图卷积网络(GCN)与长短期记忆网络(LSTM)的模型架构，通过深度融合空间特征与时序特征，实现情绪在效价、唤醒度等维度的连续识别。研究设计多组对比实验，验证了模型在DEAP等数据集上的性能优势，同时分析了不同预处理流程、模块组合及通道关联方式对识别结果的影响，为脑电情绪识别的模型优化与应用拓展提供了参考。

Abstract: This study focuses on electroencephalogram (EEG) emotion recognition. To address the low signal-to-noise ratio (SNR) and non-stationary characteristics of EEG signals, as well as the shortcomings of existing models in spatiotemporal modeling and cross-subject generalization, a model architecture fusing Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM) is proposed. By deeply integrating spatial and temporal features, the model achieves continuous recognition of emotions in dimensions such as valence and arousal. Multiple sets of comparative experiments are designed to verify the performance advantages of the proposed model on datasets like DEAP. Meanwhile, the study analyzes the impacts of different preprocessing pipelines, module combinations, and channel correlation methods on recognition results, providing references for model optimization and application expansion of EEG emotion recognition.