Fused attention mechanisms-based facial expression recognition

In recent years, the utilization of Facial Expression Recognition (FER) has gained significant traction across diverse domains including healthcare, marketing, and human-computer interface. This trend has prompted a gradual shift towards the adoption of deep learning methodologies for facial expression recognition. The paper highlights the importance of FER systems in comprehending human communication, detailing traditional methods and recent advancements in the field. It proposes a new framework integrating attention mechanisms into the VGGNet model for FER, utilizing the FER-2013 dataset for evaluation. Data augmentation techniques are employed to enhance model robustness, and attention mechanisms are integrated into the VGGNet-SE model. Through Comparative experiments, it is shown that the model of VGGNet-SE-last which adds attention mechanisms after the last convolutional layer of VGG16 achieves the most significant improvement in accuracy and reduction in Top 5 error, particularly with a ratio of 16 in SENet. Experimental results indicate that after incorporating attention mechanisms, the model can more accurately focus on important regions related to emotions in the images, leading to improved performance in recognizing facial expressions. In the future, further research on FER will be implemented by suggesting further optimizations in model architecture and exploration of more complex datasets for future research in FER.