Research on Facial Expression Recognition Based on Improved YOLOv8

To address the issues of low accuracy and slow convergence in facial expression recognition in complex scenarios, this paper proposes a facial expression recognition algorithm based on an improved YOLOv8 network. First, the upsampling operator in the YOLOv8 network is replaced with the CARAFE module. Through a content-aware mechanism, contextual information is aggregated to reduce the partial feature information loss caused by traditional upsampling methods. Second, for applications with densely populated or severely occluded objects, the inner-IOU loss function is used to dynamically scale the auxiliary box to accelerate convergence and improve generalization. Experiments on the RAF-DB facial expression dataset show that compared to the baseline model, the improved model improves Precision, Recall, and mAP@0.5 by $\mathbf{1. 9 \%}, \mathbf{2. 6 \%}$, and $\mathbf{1. 6 \%}$. Compared with other networks, the detection efficiency has advantages, and the accuracy has been greatly improved in comparative tests.