Object detection is a crucial capability for AI-driven perception in safety-critical drone and edge-vision systems, applicable in disaster response, operational security, infrastructure monitoring, and defense. Robust model performance relies on large, continuously updated datasets. However, training high-performing detectors often requires centralizing aerial imagery, which poses challenges in privacy, regulation, storage, and bandwidth. This is particularly problematic in distributed drone deployments, where visual data is generated onboard and is often impractical to transfer to centralized infrastructure.
In this work, we apply Federated Learning (FL) for object detection, allowing drones to improve a shared model while keeping image data local and private. We implemented a federated object detection pipeline using the Sherpa.ai FL platform on the KIIT-MiTA dataset and compared it with Single-drone and Centralized baselines using mean Average Precision (mAP) at IoU thresholds of 0.50 and 0.50-0.95.
Our experiments show that the proposed FL approach closely matches Centralized training while significantly outperforming Single-drone training. The best lightweight model (YOLOv6 nano), suitable for limited edge infrastructure, achieved relative gains of 52.89% and 67.80% in mAP@0.50 and mAP@0.50:0.95, respectively. These results demonstrate that FL enables scalable, high-performing, and privacy-preserving object detection across distributed drone fleets without data centralization.
Blogger's Review: This paper effectively addresses data privacy and centralization issues in drone object detection through Federated Learning, showcasing its broad potential in practical applications. The use of lightweight models offers new possibilities for edge computing devices, advancing drone technology significantly.