Emotion Recognition in Individuals with Down Syndrome: A Convolutional Neural Network-Based Algorithm Proposal

• DOI: 10.3390/sym15071435
• Publication Date: 2023-07-18T05:31:32Z
• Authors: Nancy Paredes; Eduardo Caicedo-B...; Bladimir Bacca
• Keywords: Sadness; Surprise; Hyperparameter

This research introduces an algorithm that automatically detects five primary emotions in individuals with Down syndrome: happiness, anger, sadness, surprise, and neutrality. The study was conducted a specialized institution dedicated to caring for syndrome, which allowed collecting samples uncontrolled environments capturing spontaneous emotions. Collecting through facial images strictly followed protocol approved by certified Ethics Committees Ecuador Colombia. proposed system consists of three convolutional neural networks (CNNs). first network analyzes microexpressions assessing the intensity action units associated each emotion. second utilizes transfer learning based on mini-Xception architecture, using Dataset-DS, comprising collected from syndrome as validation dataset. Finally, these two are combined CNN enhance accuracy. final processes information, resulting accuracy 85.30% emotion recognition. In addition, optimized tuning specific hyperparameters network, leading 91.48% recognition accuracy, specifically people syndrome.