A robust detect and describe framework for object recognition in early childhood education

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Educational Technology

Background:

• Preschool education is crucial for cognitive and emotional development.
• Early identification of shapes, colors, and letters builds a strong academic foundation.
• Spatial reasoning and problem-solving skills emerge from object recognition in early childhood.

Purpose of the Study:

• To present a cognitive enhancement framework using deep learning for nested object recognition in preschool education.
• To improve early childhood learning through advanced AI models.
• To develop a system for precise object recognition and semantic description.

Main Methods:

• Utilized deep learning models: You Only Look Once (YOLOv8) and Visual Geometry Group (VGG16) for object detection.
• Employed Long Short-Term Memory (LSTM) neural networks for semantic description.
• Implemented the framework in Google Colab using Ultralytics, PyTorch, and OpenCV.
• Trained and tested models on a custom dataset (PreEduDS).

Main Results:

• Achieved 94.4% accuracy for object recognition.
• Attained 96.5% accuracy for predicting precise semantic textual descriptions.
• Demonstrated widespread applicability of the cognitive enhancement framework.

Conclusions:

• The proposed framework effectively enhances preschool education and training.
• The system is well-suited for integration with augmented reality (AR) applications.
• Deep learning significantly improves cognitive development tools for young learners.