结合注意力特征融合的八度卷积表情识别方法  

Expression Recognition Method Combined with Attention Feature Fusion

作　　者：任豪 REN Hao(Guizhou Normal University,Guiyang 550025,China)

机构地区：[1]贵州师范大学物理与电子科学学院,贵州贵阳550025

出　　处：《电脑与电信》2024年第5期71-74,97,共5页Computer & Telecommunication

摘　　要：针对目前面部表情识别特征表达不足、识别精度低及参数多的问题,提出了一种结合注意力特征融合的八度卷积表情识别方法。主要创新点在于将注意力特征融合机制引入模型,优化不同尺度特征的融合;采用深度可分离网络替代传统卷积,大幅减少参数;并引入BN和PReLU提升模型稳定性和性能。实验显示,该模型在CK+和Fer2013数据集上准确率分别达98.91%和74.03%,展现了优秀的泛化能力和准确度。Aiming at the problems of insufficient expression of facial expression features,low recognition accuracy and many parameters,an octave convolutional expression recognition method combined with attention feature fusion is proposed.The main innovation point is to introduce the attention feature fusion mechanism into the model to optimize the fusion of different scale features.Deep separable network is used to replace traditional convolution,which greatly reduces parameters.BN and PReLU are introduced to improve the stability and performance of the model.Experiments show that the accuracy of the model on CK+ and Fer2013 data sets is 98.91% and 74.03%,respectively,showing excellent generalization ability and accuracy.

关 键 词：人脸表情识别 卷积神经网络 注意力特征融合机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]