A Comparative of Facial Emotion Recognition Performance Based on Cnn, Vggnet And Resnet

With the development of artificial intelligence technology, facial emotion recognition technology has become a research hotspot. The technology has been used in a variety of fields including, but not limited to, human-computer interaction, psychological research and analysis, and security monitoring. In this paper, three classical convolutional neural network models, namely Convolutional Neural Network (CNN), VGGNet and ResNet, are trained using the FER2013 dataset for the task of face emotion recognition, and their generalization ability is verified on FEC Dataset and MMA FACIAL EXPRESSION database. The experimental design covers data preprocessing, model optimization strategies, and standardized training parameter processes. The experimental results indicate that ResNet performs best in cross-dataset tests. Further analysis reveals that the residual connection mechanism effectively mitigates the overfitting problem of the deep network while enhancing the sensitivity to local expression features. The study confirms that model and structure innovation plays a key role in the extraction of complex expression features, provides a model selection basis for emotion recognition in complex scenes, and verifies the superiority of residual structure in cross-domain feature learning, which is an important reference value for dynamic emotion analysis in practical application scenarios.