Facial Expression Recognition Model Based on Improved VGGNet

Abstract

Facial expression recognition is of great significance for human-machine interaction and autonomous driving. In order to achieve simple extraction of facial emotional features and accurate classification of facial expressions, mainstream convolutional neural networks were studied, and an improved VGG-16 convolutional neural network based facial expression recognition model was proposed. This model is improved on the basis of VGG-16 network model, optimizing the number of convolution layers in the third and fourth convolution blocks, replacing the SoftMax classifier in the original VGG-16 network with the 7-tag SoftMax classifier, and replacing the original Relu activation function with the LeakeyRelu activation function. The experiment was conducted on the FER2013 dataset, and the final accuracy reached 72.42%, which is higher than the accuracy of unimproved VGGNet (66.31%) and ResNet (71.3S%). The experimental results indicate that the model has certain application capabilities in facial expression recognition.