A robust detect and describe framework for object recognition in early childhood education
Lan Lv, Suhui Yao · PeerJ Computer Science · 2025
Preschool education plays a vital role in the harmonious development of an individual. Understanding basic shapes, colors, and letters at an early age lays a strong foundation for academic excellence and emotional growth. At an early childhood stage, the skills of spatial reasoning and problem-solving can be developed by recognizing and comprehending the depicted objects. By exploring deep learning technology, this article presents a cognitive enhancement framework for recognizing nested objects. With cutting-edge models, such as You Only Look Once (YOLOv8) and Visual Geometry Group (VGG16), objects and intra-objects are detected. For semantic description, the neural network model, specifically long short-term memory (LSTM), is exploited, preceded by precise object recognition. The framework is implemented in Google Colab with the prominent packages of Ultralytics, PyTorch, and OpenCV. The models are trained and tested by a custom dataset: PreEduDS. The results of the systematic evaluation suggest that the framework has widespread applicability. A promising accuracy score of 94.4% is obtained for object recognition and 96.5% for predicting precise semantic textual description. The proposed system is well-suited for enhancing preschool education and training based on augmented reality (AR) applications.