DANA: Deep Attention Network Architecture for Facial emotions Recognition using limited resources

Abstract

Facial emotion recognition (FER) has emerged as a crucial research area with diverse applications in humancomputer interaction (HCI), customer experience analysis, and autism spectrum disorder (ASD) treatment. However, achieving high accuracy with limited resources remains challenging due to data scarcity, environmental variations, and age-related discrepancies. This research addresses these limitations by proposing a novel approach for feature extraction utilizing Convolutional Neural Networks (CNNs). CNNs excel in extracting features from images due to their inherent ability to learn complex data patterns. This research introduces a novel deep learning architecture incorporating an “intentional convolutional network” designed to prioritize key facial regions. This targeted approach aims to enhance emotion classification for seven basic emotions (contempt, anger, sadness, happiness, surprise, fear, and disgust) using the CK+48 dataset. The model surpasses existing methods, achieving a remarkable 99% accuracy on the CK+48 dataset, even with restricted computational resources. These findings contribute significantly to FER research by offering a robust and efficient solution applicable in various fields, including HCI, customer experience analysis, and ASD treatment.