YOLO11-AE: An Enhanced Fine-Grained Facial Expression Recognition Model with Dual-Channel Attention Fusion

Facial Expression Recognition (FER) plays a vital role in affective computing and human-computer interaction. However, the baseline YOLO11 model exhibits limitations in fine-grained feature extraction, which hinders its ability to capture subtle variations in facial expressions accurately. To address this challenge, we propose YOLO11-AE (Attention-Enhanced YOLO11), an improved detection network designed to enhance the representation of key expression regions. Specifically, a Squeeze-and-Excitation (SE) attention mechanism is integrated into the feature fusion stage of the C3k2 module, while an Efficient Channel Attention (ECA) module is embedded into the local feature extraction stage of the C3k module. These enhancements strengthen both the channel modeling capacity and local perceptual sensitivity of the network. Ablation studies and comparative experiments conducted on the FER2013 dataset demonstrate that the proposed YOLO11-AE significantly outperforms the original YOLO11 model in both recognition accuracy and robustness. The final results indicate a 1.6% accuracy gain, confirming the method's value in handling fine-grained features and improving expression recognition performance.