Exploring the Potential of Federated Learning CNN for Interactive Virtual Tours of UNESCO Cultural Heritage Sites: A Case Study

Cultural heritage places play a significant role in our shared history and identity, and their survival needs to be preserved and promoted. A federated learning CNN model for interactive virtual tours of culturally significant locations is presented in this paper. Utilizing the Federated Averaging technique and the MoblieNet CNN architecture, we trained the model on a simulated federated learning environment with 20 devices. Ten thousand photographs from each of the four UNESCO World Heritage Sites—Angkor Wat, Stonehenge, Machu Picchu, and Petra—comprised the total 40,000 enhanced images used to train the model—a testing set of 4,000 photos with equal representation from each site. The model's parameter values ranged from 0.93 to 0.97, yielding % overall accuracy of 95.4%. The option to customize the tour based on individual interests and preferences, which was not possible in the centralized CNN model, was appreciated by the participants utilizing the federated learning CNN approach. The participants also well-liked the model's privacy and security features, increasing confidence in the system's operation. The findings of this research show that the federated learning CNN model is very accurate and precise for item recognition and categorization in cultural heritage locations. The model's personalized and secure user interface may improve the virtual tour experience and foster system trust.