Research on Facial Expression Recognition Algorithm Based on Improved StarGAN Network

TO address the issue of overfitting in deep learning testing and pattern collapse in traditional generative adversarial networks (Gans),an based on deep learning improved Star Generative Adversarial network (StarGAN) facial expression recognition method is proposed. For this method, the regular term of pattern search and reconstruction loss are combined to improve the quality of the generated images and solve the problem of model pattern collapse. spatially separable convolution is used to extract image features to make a large number of parameters and operation cost is reduced. SVM is used for multi-task classification to improve the model generalization ability. Good recognition results were obtained on FER2013 data sets.