Deep Learning Approach for Emotion Recognition in Children with Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is a condition that affects the social interaction, communication, emotional intelligence of a children. Emotion recognition using facial expression is one of the challenging task which is making it crucial to develop an automated system. In this study we have developed an automted system which will accompany the child to understand his emotion better. We have proposed a CNN-based deep learning model to classify six emotions—joy, sadness, anger, fear, surprise, and neutral—from facial images of autistic children. We have used the ‘Autistic Children Emotions’ dataset, the model was trained and tested with augmented images to enhance generalization and robustness. The CNN architecture comprises multiple convolutional, pooling, and fully connected layers with regularization techniques like dropout and batch normalization to minimize overfitting. The model achieved a test accuracy of 40%, with "joy" being the most accurately classified emotion. Despite limitations in classifying less-represented emotions, the results indicate the potential of deep learning models to facilitate emotion recognition systems tailored for autistic children.