RA-TCN：一种多尺度时空特征融合情绪识别模型

打开文本图片集

中图分类号：TP391；R318文献标识码：A

Abstract： The dynamic evolution of emotions is embedded in multi-scale spatiotemporal patterns. However，traditional emotion recognition models are limited by their fixed receptive fields，making it difficult to effectively capture the inherent complexity，which constrains classification performance. In this paper，a multi-scale spatiotemporal emotion recognition model called RA-TCN （Receptive Field Coordinated Attention Convolutional-Temporal Convolutional Network） is proposed to improve emotion classification accuracy. Parallel temporal convolutional networks （TCN） with different dilation factors are employed to capture multi-scale temporal features，focusing simultaneously on instantaneous emotional changes， variation trends， and long-range dependencies. By integrating the Receptive Field Coordinated Attention Convolution （RFCAConv） module，key brain regions during emotional changes are localized，further enhancing performance in complex emotion recognition tasks. Experimental validation results on the DEAP and SEED datasets demonstrate that the proposed model achieves high classfication accuracy， confirming its effectiveness.

Keywords： EEG;emotion recognition;TCN;RFAConv; coordinate attention

情绪识别作为脑机接口领域的重要研究方向，在推动人机交互智能化与人性化发展方面具有显著的应用价值[1-5]。（剩余12039字）