Research on Facial Expression Recognition Based on Improved VGG19

Facial expression recognition is of great significance in the fields of computer vision and human-computer interaction. Given the problems of over-fitting, high computational complexity, and vanishing gradient in VGG19, this paper proposes an improved VGG19 network model. By introducing depth wise separable convolution, the number of parameters of the model is significantly reduced and the computational efficiency is improved. This paper applies Batch Normalization and Dropout technology to effectively alleviate the over-fitting problem and improve the model's ability for generalization. In addition, this paper introduces residual connections into the model, which solves the vanishing gradient problem and improves the training speed and model stability. This paper uses the FER2013 dataset to conduct comparative experiments on the improved VGG19 network model, the original model, and other network models. The results show that the parameter amount of the improved VGG19 decreased by approximately 45.26%, and the accuracy increased by 4.52%. In addition, the improved VGG19 model is also better than other network models implemented in this paper and has better facial expression recognition results.