Model-Driven Integration of Deep Learning for Artifact Classification in Museum Information Systems

Museum Information Systems (MIS) often rely on manual classification and keyword search, limiting accuracy and scalability. Deep learning offers a solution, but effective integration requires alignment with curatorial workflows. This study proposes a model-driven framework for integrating Convolutional Neural Networks (CNNs) into MIS to enhance artifact classification and retrieval. A prototype was built using ReactJS, Django, and TensorFlow, and it was trained on a curated subset of The Met's Open Access Images. The system employs a Hybrid-E Loss for improved classification accuracy. The model achieved 94.3% classification accuracy and real-time retrieval latency below 100 ms, with throughput exceeding 14 queries per second. The framework successfully bridges AI performance with curatorial logic, demonstrating a scalable and interpretable solution for digital heritage systems.