基于生理信号的情绪识别方法研究

Emotion recognition based on physiological signals is of considerable significance in applications such as mental health interventions and human-computer interaction. In recent years, the rapid advancement of multimodal affective computing has offered new technical support for this task. Nevertheless, the substantial inter-subject variability in physiological signals poses a major challenge, often leading to performance degradation of deep neural networks in cross-subject settings. To address this, we propose a multimodal neural architecture that combines convolutional networks with Transformer encoders. Built upon this foundation, a domain generalization strategy is further introduced to jointly enhance the model’s representational capacity and its generalizability across unseen subjects. Experiments conducted on the DEAP dataset demonstrate that the proposed approach delivers strong performance in multimodal physiological signal-based emotion recognition.