Collaborative Analysis of Multi-Regional Cultural Heritage Protection Enabled by Federated Learning

Secure and distributed information sharing has become increasingly important for intelligent monitoring systems and large-scale collaborative sensing applications associated with electromagnetic communication networks. To address data fragmentation and privacy constraints in multi-regional cultural heritage protection, this study proposes a collaborative risk analysis and strategy evaluation framework based on federated learning. A spatiotemporal graph attention network (ST-GAT) is deployed at the client side, where a two-layer long short-term memory network captures temporal dependencies and a graph attention mechanism models interregional spatial correlations under privacy-preserving constraints. FedProx regularization and differential privacy are incorporated into federated optimization to improve robustness against non-IID data while protecting sensitive information. Furthermore, a federated Bayesian decision layer evaluates conservation strategies through posterior inference, multi-objective optimization, and Monte Carlo sampling. Experimental results demonstrate that the proposed framework achieves an RMSE of 0.113 and a risk mitigation rate of 95.45%, while maintaining stable cross-regional prediction performance and substantially reducing the success rates of membership inference and statistical attacks. The proposed approach provides an effective solution for privacy-preserving collaborative conservation and offers practical insights into secure distributed perception and intelligent information transmission for future electromagnetic communication and sensing environments.