Research on Dual-Branch CNN Model for Facial Expression Recognition with Attention Mechanism

To address the issues of insufficient collaboration between global and local features, confusion among fine-grained expressions, and weak cross-scenario generalization in facial expression recognition under natural scenarios, this paper proposes a dual-branch convolutional neural network (CNN) model integrated with an attention mechanism. The model adopts a "global-local" dual-branch structure: the global branch extracts overall facial contours and semantic information through lightweight convolutions, while the attention branch incorporates a feature calibration mechanism to focus on key expression regions such as the eyes and mouth via dynamic weight allocation, enabling accurate capture of subtle features. Additionally, a hybrid loss function is designed to jointly constrain category boundary optimization and feature distribution adjustment, alleviating the problem of insufficient learning for few-shot categories caused by sample imbalance. The model's generalization capability was validated on the FER-2013 natural scene dataset. The research findings provide an effective technical pathway for fine-grained facial expression recognition in complex scenarios, and can be applied in fields such as human-computer interaction and intelligent security.