Development of a Web-Based Deep Learning Model for Real-Time Emotion Recognition Using Facial Expressions that Leverages Convolutional Neural Networks (CNN) and Attention Mechanisms

Source abstract

The automatic recognition of human emotions from facial expressions is a critical component of affective computing, with applications spanning human-computer interaction, customer service, and mental health. While traditional methods often rely on handcrafted features, deep learning approaches, particularly Convolutional Neural Networks (CNNs), have demonstrated superior performance. However, challenges remain in model interpretability and robustness to real-world variations. To address these limitations, this study proposes a web-based deep learning model for real-time emotion recognition. An enhanced facial emotion recognition model using a deep CNN integrated with a Convolutional Block Attention Module (CBAM) and Grad-CAM explainability was developed. Using the Extended Cohn-Kanade (CK+) dataset comprising 981 images across seven emotion classes, the model was trained with rigorous preprocessing and data augmentation. Results show that the proposed model achieved 98.71% accuracy, 98.7% recall, and an F1-score of 98.9%, with perfect AUC and Average Precision scores of 1.00 for all classes. The integration of CBAM improved feature focus on salient facial regions, while Grad-CAM provided visual explanations, enhancing clinical and practical trustworthiness. The system was successfully deployed as a browser-based application, demonstrating real-time inference capabilities. This study highlights the potential of attention-enhanced deep learning models in advancing transparent and efficient emotion diagnostics for real-world deployment.