DLFER: Deep Learning-based Cascaded Approach for Facial Emotion Recognition

Abstract

Facial expression identification has garnered considerable attention in recent years owing to its extensive applicability across various domains, including human-computer interaction, market research, and healthcare. The primary objective of Facial Emotion Recognition (FER) is to correlate various facial expressions with their corresponding emotional states. Advancements in deep learning have significantly enhanced the recognition accuracy of FER technology relative to conventional approaches. This study seeks to improve the accuracy for facial emotion identification by introducing a Deep Learning Cascaded Network (DLFER) founded on the EF-FER1 and EF-FER2 architectures. The hyperparameters of the EfficientNet-B3 network were fine-tuned to improve facial expression representation and classification. The trials utilized the widely recognized Facial Expression Recognition 2013 (FER2013) dataset, comprising 35,887 greyscale photos of faces, each linked to one of seven specific emotions. The model's performance was assessed using accuracy, precision, recall, and F1 score. A comparative analysis was also performed with similar modern studies. The trials demonstrated that the Cascaded Network (DLFER) attained a classification accuracy of 82.09%, surpassing that of state-of-the-art models.