Hybrid ResNet50–ResNet34 Model for Enhanced Facial Expression Recognition

Abstract

Facial expression recognition is a fundamental challenge in computer vision and plays a crucial role in enhancing human-computer interaction. It enables the automatic interpretation of human emotions from facial images, with important applications in affective computing, social robotics, and psychological research. In this work, we propose a hybrid deep learning approach combining ResNet50 and ResNet34 models for effective facial expression classification. These models, pre-trained on large-scale image datasets, demonstrate powerful feature extraction capabilities and have achieved remarkable performance in various computer vision tasks. Our method follows a systematic approach, beginning with the collection and preprocessing of a labeled facial expression dataset. The dataset undergoes face detection, alignment, and normalization to ensure uniformity and reduce noise. The preprocessed data is then partitioned into training, validation, and testing subsets. We fine-tune the ResNet50 and ResNet34 models using transfer learning, adapting them specifically for facial expression recognition. Optimization techniques such as SGDM, ADAM, and RMSprop are applied to update model parameters and minimize the categorical cross-entropy loss function. The models are evaluated based on key performance metrics, achieving a high accuracy of 98.19% on the validation set. Finally, the system is tested on unseen facial images to verify its generalization ability. The proposed approach demonstrates robust and accurate facial expression classification, advancing the development of intelligent emotion analysis and human-computer interaction systems.