Enhanced Facial Emotion Recognition Using Transfer Learning with ResNet152

The prospective applications of facial expression-based emotion recognition have sparked a lot of interest in domains like camera technology, mental health analysis, and human-computer interaction. Using the ResNet152 architecture, we suggest a unique facial emotion recognition (FER) technique that makes use of preprocessing, data augmentation, and transfer learning. While deep learning techniques, particularly when combined with CNNs, are known to be successful for classification problems, conventional methods have difficulty optimizing both feature extraction and classification at the same time. Through extensive testing on a dataset, the research shows that the suggested method is effective, with ResNet152 obtaining an accuracy of $\mathbf{8 4 \%}$. Experiments on randomly selected photos from different sources revealed encouraging outcomes for precise and instantaneous FER. This study presents a strong method for FER that makes use of transfer learning and deep learning, providing notable improvements over traditional methods.