Leveraging Attention Mechanisms in ResNet101 for Enhanced Facial Emotion Recognition

Abstract

Facial emotion recognition serves many real-world applications across health care and security and human-computer interaction because these fields require advanced models recognizing emotions under true practical situations. The proposed Advanced Facial Emotion Recognition (FER) system utilizes ResNet-101 with Attention Mechanism to resolve gaps found in conventional methods whose examples include: ResNet50 and DeepExpr together with Ensemble Multi-feature approaches and both Adversarial neural networks and Attentional CNNs. The proposed framework demonstrate limited capability to recognize fine-grained embodied perceptions including emotions as well as faces and views together with novel realistic environmental changes and obscuring effects. The proposed model detects five complex emotions (sadness, anger, happiness, disgust and neutrality) by giving specific attention to the targeted emotion regions in facial areas. The proposed system reports F1 scores between 0.92 to 0.95 alongside evaluations demonstrating precision ranging from 0.88 to 0.97 and recall ranging from 0.92 to 0.95 for all emotions. This method demonstrates 99.6% accuracy in recognition yet its Kappa score stands at 94.1 because of slight discrepancies compared to classic models.