Transfer Learning Based Facial Emotion Recognition

The efficiency and accuracy of facial emotion detection imparts on mental health monitoring through Transfer Learning with Facial Emotion Recognition (FER). The Transfer Learning has a resource-efficient and robust solution through FER which has advancing capabilities and FER categorises human emotions through the facial expressions in the images or video sequences. This study explores an advanced FER methodology by utilizing the ResNet-Convolutional Block Attention Module, deep Convolutional Neural Network, known for its depth and enhanced attention mechanisms. The model was pretrained on ImageNet, was fine-tuned to classify facial emotions using the FER + dataset, augmented with preprocessing techniques to enhance robustness and accuracy. A multi-step methodology, starting with data augmentation employed to diversify the training data and overcome overfitting. The study focuses on the ResNet-CBAM architecture, rigorously fine-tuning its hyperparameters and experimenting with the Adam optimizer method. The model achieves a single-network accuracy of 83.74% on FER + dataset. The findings suggest that the model, with appropriate fine-tuning, can serve as a powerful tool for emotion recognition applications. This research contributes to the growing body of knowledge on deep learning for FER, offering insights into model optimization. The proposed method offers a scalable and efficient solution suitable for real-time emotion recognition, with potential applications in affective computing, psychological research. This study demonstrates the effectiveness of ResNet-CBAM in FER, highlighting its potential for various real-world applications.