This paper develops a cutting-edge multimodal federated learning framework, integrated with distributed ledger technologies, designed specifically for UAV delivery scenarios. The framework adopts various data modalities, including user pictures, behavior, and location, to dynamically optimize delivery routes and schedules, thus enhancing both user privacy and security of the delivery process. By employing federated learning, this framework allows data to be processed locally on individual devices, significantly enhancing both user privacy and data integrity. The integration of distributed ledger technology ensures that all updates to the federated model are not only immutable and traceable, but also secure. Through comprehensive evaluations, our framework shows outstanding improvements in both the efficiency and security of UAV deliveries. These findings show the transformative potential of our approach to establish user-centric, efficient, and secured UAV delivery systems.