همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Enhancing Facial Emotion Recognition Using YOLO11 Classification on the AffectNet Dataset

نویسندگان :

Reza Nasiri¹ Seyed Enayat Alavi² Mohammad Javad Rashti³

1- دانشگاه شهید چمران اهواز 2- دانشگاه شهید چمران اهواز 3- دانشگاه شهید چمران اهواز

کلمات کلیدی :

Facial Expression Recognition،YOLO11،AffectNet،Deep Learning،Affective Computing،Convolutional Neural Networks.

چکیده :

Facial Expression Recognition (FER) is a core component of affective computing, enabling intelligent systems to interpret human emotional states in real-world environments. Despite considerable progress, FER performance on large-scale in-the-wild datasets such as AffectNet remains limited due to illumination variations, pose changes, occlusions, intra-class variability, and annotation noise. Existing architectures including attention-enhanced CNNs, landmark-guided models, and transformer-based approaches have achieved incremental improvements but generally remain below 70% accuracy on the AffectNet-7 benchmark. In this study, we introduce a comprehensive FER framework based on YOLO11- Classification, a high-capacity classification architecture with 28.3 million parameters, optimized feature extraction, and enhanced receptive-field modeling. Without relying on handcrafted FER-specific modules such as facial alignment, landmark extraction, or region-attention engineering, the proposed system achieves 79.50% Top-1 accuracy on AffectNet-7, surpassing the previous state-of-the-art (NorFace, 68.69%) by an absolute margin of 10.81%. Extensive experiments demonstrate the robustness of YOLO11-Classification under real-world challenges, including occlusions, extreme head poses, and class imbalance, while confusion matrix analysis confirms consistent improvements across all emotional categories, particularly those with limited representation. These findings establish a new state-of-the-art for AffectNet-7 and highlight the potential of high-capacity general-purpose architectures as powerful alternatives to FER-specific handcrafted designs.