With the acceleration of the process of educational informatization, the traditional campus management model faces core problems such as serious data silos, low efficiency of cross-departmental collaboration, and insufficient dynamic resource allocation. In particular, there are significant bottlenecks in multi-source heterogeneous data fusion and real-time service response. In response to the above challenges, this study proposes a smart campus construction framework based on multimodal data fusion and intelligent decision optimization. First, a cross-system data collaboration mechanism based on federated learning is designed. Through the Dynamic Feature Alignment (DFA) algorithm, secure sharing and semantic mapping of heterogeneous data such as academic affairs, security, and energy consumption are achieved, solving the problems of privacy leakage and high adaptation costs. Next, a lightweight edge computing node cluster is introduced, and an improved spatio-temporal attention network (Spatio-Temporal Transformer, ST-Transformer) is used to perform distributed analysis of real-time data such as campus traffic and equipment status. Finally, a knowledge graph-driven resource scheduling engine is built, and deep reinforcement learning (Deep Recurrent Q-Network, DRQN) is combined to dynamically optimize the reservation strategies of classrooms, laboratories and other places to maximize resource utilization while meeting multiple constraints. Experiments show that the cross-departmental collaboration mechanism based on federated learning significantly improves business response efficiency and reduces the risk of data privacy leakage. The ST-Transformer network can effectively schedule resources, especially during peak periods of resource demand, and improves resource utilization, with overall resource utilization increased by about 17%. Compared with the traditional static strategy, the optimization strategy based on DRQN reduces the resource conflict rate and improves the resource allocation efficiency. This paper provides an effective technical solution for smart campus resource management, and has important theoretical significance and application prospects.