MD-EPN: An Efficient Multimodal Emotion Recognition Method Based on Multi-Dimensional Feature Fusion

Multimodal emotion recognition is extensively applied in diverse areas including driving monitoring, online education, telemedicine, and customer service, marking it as a significant technology in contemporary research. Traditional emotion recognition methods predominantly concentrate on unimodal data processing, which can lead to low accuracy and substantial information loss. In response to these issues, this paper introduces an innovative Multi-Dimensional Emotion Perception Network (MD-EPN) that leverages advanced technologies to integrate textual, auditory, and visual data, thereby aiming to enhance the overall comprehensiveness and precision of emotion recognition. By focusing optimizations on crucial dimensions, the proposed architecture not only bolsters the model's performance but also strikes an effective balance between computational efficiency and accuracy retention, exhibiting exemplary emotion classification accuracy on the CMU-MOSI dataset.