FER-Net: Leveraging Attention Mechanisms in EfficintNetB5 model for Enhanced Emotion Recognition in Augmented Images

Facial emotion recognition is a critical component in the development of advanced human-computer interaction systems, with applications spanning security, healthcare, and social robotics. This study explores the efficacy of advanced deep learning models, specifically EfficientNetB5 model in identifying emotions from facial images. Leveraging with augmented data, our approach integrates attention mechanisms, batch normalization, convolutional layers, global average pooling layers, and dropout techniques to enhance model performance. The proposed model demonstrated significant performance accuracy of 87.5%, highlighting the importance of emphasizing relevant image regions for better emotion detection. Additionally, we conducted a comparative analysis with other prominent deep learning architectures, including AlexNet, XceptionNet, ResNet, MobileNet and EfficientNetV2M and proved the superiority of the proposed model against them.