Abstract
<title>Abstract</title> <p>Facial expression recognition in children is a challenging problem due to limited data, domain variability, and computational constraints. Existing approaches often rely on deep neural networks that achieve high accuracy but require significant computational resources, making them less suitable for real-time applications. In this work, we propose a lightweight diffusion-augmented framework for efficient children facial expression recognition. The framework integrates a lightweight convolutional neural network for feature extraction, diffusion-based data augmentation to address data scarcity and domain variability, and knowledge distillation to improve model efficiency. The proposed approach focuses on achieving a balance between recognition performance and computational complexity. Experiments are conducted on two benchmark datasets, LIRIS-CSE and CAFE. The results show that the proposed method outperforms baseline models such as ResNet-50, MobileNet, and LITE-FER in terms of accuracy, precision, recall, and F1-score, while maintaining lower computational cost. These results demonstrate that the proposed framework is suitable for real-time and resource-constrained applications in education, healthcare, and assistive systems.</p>