FerNeXt: Facial Expression Recognition Using ConvNeXt with Channel Attention

Facial expression recognition has contributed significantly to various domains of life from healthcare and education to marketing and sales. This has led to extensive research in trying to improve recognition methods using deep learning. These methods aim to integrate advanced feature extraction techniques with enhanced classification accuracy. In this paper, a method for facial expression detection called (FerNeXt) was proposed which is based on the already existing ConvNeXt network architecture. An Efficient Channel Attention (ECA) block was introduced within the ConvNeXt architecture to enhance the model's attention to important channel features. Additionally, an Affinity loss function was incorporated to optimize class separability. The method was experimented on two large datasets: AffectNet and RAF-DB to show the proposed technique's capability. The results obtained show that the proposed model demonstrates good performance, consistently achieving higher accuracy and outperforming state of the art facial expression recognition approaches.