Abstract: In view of the problems of traditional facial expression recognition algorithms, such as poor feature extraction ability, low recognition rate and high misclassification rate, we proposed an improved facial expression recognition algorithm based on generative adversarial network (GAN). Using the game theory of generating antagonism network, the feature extractor, feature combiner and discriminator were designed respectively. Through the adversarial training between the discriminator and feature extractor, the feature extraction ability of the feature extractor and the accuracy of the classifier for facial expression recognition were continuously enhanced, and it was applied to the intelligent monitoring of working state of staff. The working state was judged according to the results of expression recognition, so as to reasonably allocate the laboratory resources and improve the utilization rate of laboratory resources. The several experimental results of the improved algorithm on the CK+ dataset show that the proposed algorithm has higher robustness and can effectively improve the rate of facial expression recognition.

Key words:  expression recognition, generative adversarial network (GAN); convolutional neural network, deep learning, intelligent monitoring of working state