AI framework uses Swin Transformer and federated learning for private UAV image transmission

A team of eight researchers from multiple institutions has introduced a novel framework for privacy-preserving image transmission in low-altitude networks, targeting the growing UAV applications in logistics, inspection, and emergency response. Their proposed Swin Transformer-based Semantic Communication (STSC) architecture extracts multi-scale semantic features under constrained bandwidth conditions, addressing the dual challenge of limited transmission capacity and stringent data privacy requirements.

The framework leverages federated learning to enable global model training across distributed UAVs without sharing raw data, thus preserving user privacy. Dedicated communication and computing nodes deployed on UAVs enhance real-time coverage and flexibility. In simulation experiments on CIFAR-10, STSC achieves a 5.7dB improvement in Peak Signal-to-Noise Ratio (PSNR) compared to DeepJSCC baselines, with superior convergence and generalization. This work offers a practical solution for efficient, secure image transmission in bandwidth-constrained low-altitude networks, paving the way for more reliable and private UAV operations.