An IoT-Based Architecture Exploiting Deep Learning Approaches for the Maintenance of Cultural Heritage

Preserving artistic and cultural heritage is a fundamental need that requires the use of new technologies to support experts in the field. In particular, the possibility of exploiting the Internet of Things paradigm allows the acquisition of data that, when properly processed, enables the prediction of potential threats and, therefore, the prevention of cultural artifacts from being damaged by weather, atmospheric conditions, or pollution. To this end, this work introduces an architecture that exploits two different processing strategies starting from the acquired data. The first exploits the data to detect potential anomalies using AuotEncoders. The second exploits physical models that are analyzed using Physics-Informed Neural Networks. These analysis strategies allow simulations and alerts to be obtained to support artistic and cultural heritage conservation experts. The experimental phase aims to evaluate both approaches and obtain satisfactory results.