J. Eur. Opt. Society-Rapid Publ. 2026, 22, 49 Ó The Author( s), published by EDP Sciences, 2026 https:// doi. org / 10.1051 / jeos / 2026043 Available online at: https:// jeos. edpsciences. org
Journal of the European Optical Society-Rapid Publications
RESEARCH ARTICLE
SynthSwarm: A controllable synthetic dataset for UAV Swarm detection
Chaowen Zheng, Limin Liu, Luyi Zhang, Haojie Yang, Jianyu Liu, Qiang Fu, Qing Yang *, and Xiwei Guo * Shijiazhuang Campus, Army Engineering University of PLA, Shijiazhuang 050003, PR China Received 22 February 2026 / Accepted 7 May 2026
Abstract. Reliable detection of unmanned aerial vehicle( UAV) swarms is essential for airspace security and defense applications, however the scarcity of large-scale, densely annotated training data remains a critical bottleneck. Collecting real-world swarm data is costly, logistically challenging, and constrained by airspace regulations, while manual annotation of numerous small, fast-moving targets is time-consuming and prone to errors. To address these challenges, this paper presents SynthSwarm, a large-scale synthetic dataset specifically designed for UAV swarm detection in long-range aerial surveillance scenarios. The dataset is generated through a controllable simulation pipeline built on the Unity engine, enabling precise six-degree-of-freedom( 6-DoF) pose specification for each UAV instance and automatic pixel-accurate bounding box annotation without manual labeling. SynthSwarm comprises 7000 high-resolution images( 1920 1080) containing 31,542 UAV instances, with systematic variations in swarm density, formation patterns, target scale, and environmental conditions. Statistical analysis reveals that 67.3 % of the targets qualify as small objects, reflecting the inherent difficulty of detecting distant UAV swarms. We benchmark several representative deep learning detectors, including one-stage detectors( YOLOX, YOLOv6, YOLOv12, YOLOv13), the two-stage detector Faster R-CNN, and the Transformer-based detector RT-DETR. Experimental results demonstrate that the dataset poses significant challenges for existing methods, particularly in high-density and small-target scenarios. Furthermore, cross-dataset on the MMFW-UAV dataset experiments validate the effectiveness of synthetic data as a pre-training source for improving detection performance on real UAV datasets. The dataset and generation pipeline are publicly available to facilitate further research in UAV swarm detection.
Keywords: Synthetic dataset, UAV swarm detection, Small target detection, Deep learning.
1 Introduction
Unmanned aerial vehicle( UAV) swarms have recently emerged as a key enabling technology in both civilian and defense sectors due to their high flexibility, scalability, and robustness [ 1 ]. Cooperative multi-UAV systems are increasingly deployed for applications such as large-area environmental monitoring [ 2 ], precision agriculture [ 3 ], search and rescue [ 4 ], and outdoor firefighting [ 5 ]. Compared to single-UAV scenarios, swarm operations involve more complex spatial formations, higher target densities, and more dynamic interactions among agents, significantly increasing the challenges of perception and situational awareness. In particular, reliably detecting multiple small UAVs in cluttered environments remains a difficult task [ 6 ]. Beyond these perception challenges, the lack of highquality training data remains a critical bottleneck limiting
* Corresponding authors: Qing Yang, yang _ qing @ aeu. edu. cn; Xiwei Guo, hep0168 @ aeu. edu. cn. the robustness of data-driven detection algorithms for drone swarms. Publicly available UAV detection datasets either focus on single-UAV or small-scale multi-UAV scenes or contain limited variations in swarm size, formation patterns, and environmental conditions. For instance, multisensor and multi-view datasets such as MMFW-UAV [ 7 ] provide valuable resources for air-to-air vision tasks involving fixed-wing UAVs, but they mainly target single-platform perception and do not explicitly model swarm formations. Other works have also highlighted the limitations of existing datasets in addressing dense multi-UAV scenarios [ 8 ]. In real deployments, the 3D poses and spatial formations of UAV swarms are difficult to precisely control on demand, further hindering the systematic acquisition of diverse training examples, especially for dense or safety-critical configurations. Collecting real swarm UAV data at scale is expensive and logistically challenging, often subject to strict airspace regulations and safety considerations. Moreover, accurately annotating numerous small, fast-moving UAV targets in high-resolution videos is extremely
This is an Open Access article distributed under the terms of the Creative Commons Attribution License( https:// creativecommons. org / licenses / by / 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.