Federated learning for UAV surveillance with camera vision and radar point cloud data fusion to enhance sensing-assisted communication

With the rapid expansion of the low-altitude economy, Unmanned Aerial Vehicles (UAVs) have been increasingly adopted for a wide range of civilian applications. Nevertheless, their large-scale deployment imposes substantial challenges on existing surveillance and communication infrastructures. This paper presents a Federated Learning (FL)-based UAV surveillance framework that fuses camera vision and radar point cloud data to enhance sensing-assisted communication. Each client performs local model training with multi-modal data, and the server aggregates parameters to build a privacy-preserving global model. A lightweight multi-modal neural network jointly processes radar and vision data to achieve efficient UAV detection. Experiments on the DeepSense dataset and a real-world testbed achieve average precisions of 90.41% and 97.95%, respectively. Communication simulations further demonstrate an average channel capacity improvement of up to 6% over camera-only baselines, confirming that the proposed FL-based multi-modal fusion framework enhances UAV detection accuracy and communication efficiency while offering a scalable solution for future intelligent low-altitude network systems.