An Analysis of Facial Expression Recognition Based on the EfficientNet - B4 Model

Abstract

Facial expressions are important elements in interpersonal communication. When it comes to human - computer interaction, accurately recognizing users’ facial expressions can more effectively understand their emotional tendencies, help to further seize their needs, and give responses that better satisfy their expectations. Making use of deep learning based methods for facial expression identification can play an important role in domains like healthcare and criminal investigation. The study puts forward the idea of using a facial expression recognition model based on EfficientNet - B4. It will be trained with a further improved AffectNet dataset, and the trained model’s skill at recognizing test expressions will be evaluated to judge training efficacy. Experimental results show that the current model does an excellent job in classification on the AffectNet dataset. The accuracy of the model improves steadily, and the difference in accuracy between the test and training sets is less than 0. 05.The experiment proves that the EfficientNet - B4 model has a high capacity to deal with the 8 - class emotion classification task of the AffectNet dataset, showing its competence in image recognition and classification. The procedures and conclusions of this study are not only useful as references for related research but also provide valuable ideas for subsequent model refinement and dataset improvement.