Developing an Image-Based Model for Identifying Learning Devices

Abstract

Manually tracking and identifying learning devices remains a significant challenge in educational settings, making effective classroom resource management difficult, especially in developing countries where digital infrastructure is often constrained. This study addresses this problem by utilising deep learning techniques to develop an automated image-based model for device identification. The suggested approach utilises a convolutional neural network (CNN) architecture, incorporating data augmentation to enhance dataset diversity, transfer learning with pre-trained models (such as ResNet-50), and fine-tuning for improved generalisation across a range of real-world scenarios. Training and evaluation were conducted using a carefully selected dataset of 10,000 photos that included a variety of devices in various lighting and occlusion conditions. With near real-time processing speeds of 30-50 ms per image, the model achieved a classification accuracy of 92.5% (p < 0.01) on standard images and 83.7% (p < 0.05) on partially occluded or noisy inputs with 10-25% distortion. These outcomes underscore the model's resilience and utility in dynamic learning environments, particularly for educational institutions in developing regions. By tailoring computer vision techniques for educational resource management, this study presents a novel approach with substantial potential for uses such as usage analytics, device tracking, and loss prevention. Specifically, this model offers tangible benefits for educational resource management, including enhanced device tracking, valuable usage analytics, and robust loss prevention, thereby directly contributing to improved teaching and learning outcomes. This work advances AI-driven tools in education by offering a scalable and effective solution, opening the door for broader adoption in academic and industrial settings.