Research on face expression recognition based on the ResNet network
Liuchang Shen · 2024
Currently, face expression recognition suffers from the problem of significant differences in recognition accuracy of different expression categories and weak generalization ability; in order to improve the recognition accuracy, this paper proposes a face expression recognition method based on ResNet50-PMCA network, and at the same time completes the balanced processing of the dataset. Firstly, high-quality and balanced expression data samples are generated through the ADA-StyleGAN2 network to expand the training set effectively. Secondly, the Pyramidal Multidimensional Collaborative Attention Module (PMCA) is introduced into the ResNet50 network to improve the ability to recognize model expressions. Compared with the original ResNet50 network, the improved ResNet50-PMCA network improves the recognition rate on the equalized B-CK+ dataset by $\mathbf{4 . 4 1 \%}$. The experimental results show that this method significantly improves the recognition rate when background clutter, and unbalanced expression categories in the natural environment.