Research on Facial Expression Recognition Based on Improved VGGNet

Abstract

Based on VGGNet for facial expression recognition, this paper compares the training test results of three different structures of VGNet16-C, VGGNet16-D, VGGNet19 on FER2013 face expression data set, among which the highest accuracy (68.475%) of expression VGGNet16-D network model for performance improvement. In order to promote network training to prevent over-fitting and enhance the generalization capability, this paper uses the optimized policy of batch normalization strategy, cross-entropy loss function, stochastic inactive, and data entry enhancement to scale up the network model accuracy toward expression recognition by 3.909%, and 1.223% higher than the champion preferable mode in ICML2013. In this paper, the performance improvement method of VGGNet-D facial expression recognition with joint optimization strategy is effective.