文献类型：期刊文献

英文题名：AT-PMF: Progressive multi-modal fusion with adversarial training for physiological emotion recognition

作者：Li, Dongdong[1];Huang, Shengyao[1];Wang, Zhe[1]

机构：[1]East China Univ Sci & Technol, Shanghai 200237, Peoples R China

年份：2026

卷号：172

外文期刊名：PATTERN RECOGNITION

收录：;EI(收录号：20254619517807);WOS:【SCI-EXPANDED(收录号:WOS:001623256900005)】；

语种：英文

外文关键词：Emotion recognition; Multi-modal fusion; Physiological signals; Adversarial training

摘要：Emotion recognition via physiological signals is valued for its objectivity and real-time capabilities. Electroencephalogram (EEG) signals provide direct brain activity insights but face challenges like noise and limited spatial resolution. To address these limitations, this study introduces a novel adversarial training based progressive multi-modal fusion framework (AT-PMF) for physiological emotion recognition. The framework first extracts modality-specific feature, where a temporal convolution and channel-enhancer module is designed for peripheral physiological signals, and a temporal convolution and spatial attention module is proposed for EEG signals. Adversarial training is then applied to disentangle modality-complementary and modality-consistent features, with a focus on prioritizing EEG as the dominant modality. A progressive multi-modal fusion strategy is then used to incrementally integrate these features, ensuring a robust and discriminative representation of emotional states. Extensive experiments on the DEAP and MAHNOB-HCI datasets demonstrate that the AT-PMF framework outperforms existing methods in both recognition accuracy and parameter efficiency.