Facial Expression Recognition Model Based on Improved Deep Residual Network

In recent years, feature-based attention mechanisms (including channel attention and spatial attention) have been widely used in computer vision. Our research mainly explores the improvement effect of Coordinate Attention on the key regions of model attention in facial expression classification tasks, and the impact of its application in different positions in the backbone network of the model on the accuracy of the model. We use ResNet-18 as the base model, and by introducing Coordinate Attention into the network structure, Two improved models, ResNet18+CA-attention (layer1) and ResNet18+CA-attention (Layer1-2-3), are designed. We train and test our proposed model on the FER-2013 dataset, and the results show that the improved model is superior to the original model in multiple performance indicators. Through the Grad-CAM visual analysis, it is verified that the improved model significantly improves the attention to the key facial regions, and enhances the accuracy and interpretability of classification.