Attention-Enhanced Convolutional Neural Network for Emotion Recognition in Facial Expressions

For the need for efficiently detecting the facial emotions is essential for a wide array of application, including human-computer-interaction, automated physiological analysis and interactive system in fields like health care, education and customer care services. Recognizing the needs of the upcoming future our project aims to build a modern and advance facial emotion recognition system using the modern AI techniques using the Keras and OpenCV. The primary contribution of this study was in creating and using a generalized data set by integration of several well-established datasets including CK +, ExpW-F, FER-2013, KDEF, RAF-DB, and SFEW2.0. The data preprocessing is firstly done manually and then automatically by the model to eliminate the corrupted files and carefully check the quality and correcting mislabeled images to increase data consistency and reduce the training errors.to address the feature fusion we integrated the attention feature fusion (AFF) block which allows sensitive feature forms various layers to improve the model sensitivity to collect the emotion clues. We also added the custom data generator for the augmentation of data and the training protocol includes the suite of adaptive learning rate techniques and early stopping techniques, optimized for high performance and minimal overfitting the experiment result demonstrates the model substantial accuracy, improvements and adaptability with the help of generalized dataset the model is trained to work in real-world emotion expression.