无线电工程2025，Vol.55Issue(8)：1590-1597,8.DOI:10.3969/j.issn.1003-3106.2025.08.005
基于残差注意力和BiLSTM的视听情绪识别
Audiovisual Emotion Recognition Based on Residual Attention and BiLSTM
摘要
Abstract
Despite the excellent performance of AI systems on cognitive tasks,their applications in human-computer interaction scenarios still have significant limitations due to the lack of human emotion comprehension,making emotion recognition research a key component in improving machines'adaptability to the environment.To solve the problem of poor emotion classification caused by insufficient multimodal fusion,an emotion recognition solution based on audiovisual fusion is proposed.Facial features are extracted using the MobileNet model,while audio features are extracted using Emotion2vec,and the two modal features are respectively processed by Bi-dictional Long Short-Term Memory(BiLSTM)to combine the front and back frame features,and the residual attention module is used to fuse the two modal features of audio and vision.The accuracy of emotion recognition in the experiment on the public dataset RAVDESS is 91.33%and the comparison with other methods shows that this method significantly improves the accuracy of audiovisual emotion recognition.关键词
注意力机制/情绪识别/神经网络/视听融合Key words
attention mechanism/emotion recognition/neural network/audiovisual fusion分类
信息技术与安全科学引用本文复制引用
刘充,江雪..基于残差注意力和BiLSTM的视听情绪识别[J].无线电工程,2025,55(8):1590-1597,8.基金项目
江苏省重点研发计划(BE2023087) （BE2023087）
南京邮电大学江苏省通信与网络技术工程研究中心开放课题资助(JSGCZX23007) Key Project of Natural Science Foundation of Jiangsu Province(BE2023087) （JSGCZX23007）
The Open Project Fund of Jiangsu Engineering Re-search Center of Communication and Network Technology(JSGCZX23007) （JSGCZX23007）
