Federated Learning for Object Detection: Enabling Collaborative Drone Learning Without Centralizing Data

This paper applies federated learning to object detection across distributed drones, using the Sherpa.ai platform and the KIIT-MiTA dataset. It compares federated training with single-drone and centralized baselines using mAP@0.50 and mAP@0.50:0.95. The abstract reports that the federated approach remains close to centralized training while substantially outperforming single-drone training. The best lightweight model, YOLO26 nano, achieved relative gains of 52.89% and 67.80% over single-drone training at the two evaluation settings, respectively. The proposed setup keeps image data on local drones, targeting privacy, bandwidth, and deployment constraints.