We propose a multimodal learning framework that integrates wireless channel state information with UAV flight trajectories to enhance the stability and security of physical layer key generation using deep learning models. The proposed technique effectively defends against future quantum decryption threats and is particularly suitable for environments with limited computational resources and the absence of trusted third-party authentication, thereby strengthening data protection in UAV networks.

This technology combines wireless channel state information (CSI) with UAV flight data and utilizes deep convolutional neural networks to enhance channel reciprocity, enabling key generation in dynamic flight environments. Compared to existing methods, it improves reciprocity by over 30%, reduces computational resources by up to 1000 times, and lowers energy consumption by nearly 100 times, establishing an innovative foundation for UAV communication security.

Our team＇s SkyCrypt technology combines the physical characteristics of wireless channels with drone kinematic parameters to generate dynamic keys, enhancing security. In military, logistics, surveillance, and other fields, SkyCrypt can resist quantum computing attacks, prevent data tampering, and ensure stable system operation. In the future, it is expected to be widely used in smart cities and highly automated systems, providing secure communication solutions.