Multiple Face-Emotion Recognition Using Attention Mechanisms in Deep Learning

Emotions were conveyed through facial expressions and could represent unspoken words. While humans could easily interpret these expressions, it remained a challenging task for machines. Recent research in Facial Emotion Recognition (FER) had been dominated by two major concerns: biases associated with emotion-related factors such as lighting effects, head posture, and individual uniqueness, and overfitting caused by insufficient training data. These challenges led to innovations such as integrating FER into robotic systems to enable emotional expression in robots. This study explored the application of deep learning in FER using a facial-emotion-recognition dataset comprising eight target emotions: disgust, anger, sadness, surprise, neutrality, happiness, fear, and contempt. Due to variations in lighting conditions and head positions, it was very hard for computers to identify emotions in natural settings. To address these challenges, this paper proposed a fine-tuned Transfer Learning approach using a Convolutional Neural Network (CNN) based on the EfficientNet-B0 pre-trained model. Custom classification layers and feature extraction techniques were incorporated to enhance image categorization. The proposed model achieved outstanding performance, attaining an accuracy of 98 %, surpassing previous works.