Kids' Emotion Recognition Using Various Deep-Learning Models with Explainable AI.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Facial expressions provide crucial social cues for understanding human emotions and intentions.
• The rise of online education necessitates improved interactive tools, especially for children.
• Recognizing children's emotions is challenging due to developmental differences in facial expressions compared to adults.

Purpose of the Study:

• To develop an explainable artificial intelligence (XAI) model for accurate emotion recognition in children using visual cues.
• To address the limitations of existing datasets and models for children's emotion recognition.
• To enhance the interactivity and effectiveness of online learning platforms for children.

Main Methods:

• Utilized two datasets: LIRIS Children Spontaneous Facial Expression Video Database and a novel author-created dataset.
• Created mesh versions (LIRIS-Mesh, Authors-Mesh) using 3D 468 landmark points to capture distinct facial structures.
• Compared seven Convolutional Neural Network (CNN) models and employed XAI techniques (Grad-CAM, Grad-CAM++, SoftGrad) for model interpretability.

Main Results:

• Achieved the highest accuracy of 90.98% on the author-created dataset and 89.31% on the LIRIS dataset.
• Demonstrated superior performance compared to previous studies on the LIRIS dataset (75% accuracy).
• XAI methods successfully localized facial features contributing to specific emotion classifications.

Conclusions:

• Children's facial expressions differ significantly from adults', requiring specialized recognition models.
• The proposed XAI-based emotion recognition system offers high accuracy and interpretability for children's online learning.
• This research provides a foundation for developing more engaging and effective AI-driven educational technologies for children.