Empirical Predeployment Assessment of Image-Degradation Robustness, False Alarms, and FP32 On-Device Execution for YOLO-Based Drone Detectors in Counter-UAS Surveillance
Drones · 06.11.2026 · 10 (11) · 829
DOI: 10.3390/drones10110829 ↗Kurz gefasst
Fachartikel in der Zeitschrift Drones, online erschienen am 06.11.2026. Verfasst von Inkyoun Lee und 1 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Vision-based counter-unmanned aircraft system (C-UAS) surveillance requires evidence on detection under degraded imagery, false alarms, deployment fidelity, and edge execution. We present an empirical predeployment assessment of YOLO26s, YOLOv12s, and YOLOv13s on a scenario-group-disjoint VisioDECT split (19,417 images; 3424 test images), using five independently trained seeds per model and six controlled image quality stressors. Clean mAP50–95 was 0.4831, 0.4599, and 0.4300, respectively, whereas mean retention under the stressors was 71.60%, 64.05%, and 66.15%; thus, the clean and robustness rankings differed. For seed 42, salt-and-pepper noise produced a monotonic dose response. At 2% noise, mAP50–95 fell to 0.1144, 0.0025, and 0.0012, with model-specific false-positive and false-negative failure modes, and a 3 × 3 median filter restored it to 0.5056, 0.4474, and 0.4200. On 600 auxiliary COCO val2017 negative images, false positives per image at confidence 0.25 were 1.345, 0.868, and 0.578. Nine matched FP32 PyTorch–ONNX pairs differed by at most 1.09 × 10−6 in mAP50–95, and Android application accuracy stayed within 0.19 percentage points of desktop references, whereas the tested FP32 1280 × 1280 camera stream configuration remained below one frame per second. These results support multi-dimensional predeployment evaluation of the vision layer in C-UAS surveillance.
Zitieren
APA
Lee, I., & Kim, Y. (2026). Empirical Predeployment Assessment of Image-Degradation Robustness, False Alarms, and FP32 On-Device Execution for YOLO-Based Drone Detectors in Counter-UAS Surveillance. Drones, 10(11), 829. https://doi.org/10.3390/drones10110829
Vancouver
Lee I, Kim Y. Empirical Predeployment Assessment of Image-Degradation Robustness, False Alarms, and FP32 On-Device Execution for YOLO-Based Drone Detectors in Counter-UAS Surveillance. Drones. 2026;10(11):829. doi:10.3390/drones10110829
BibTeX
@article{Lee2026,
author = {Lee, Inkyoun and Kim, Young-Kuk},
title = {Empirical Predeployment Assessment of Image-Degradation Robustness, False Alarms, and FP32 On-Device Execution for YOLO-Based Drone Detectors in Counter-UAS Surveillance},
journal = {Drones},
year = {2026},
volume = {10},
number = {11},
pages = {829},
doi = {10.3390/drones10110829}
}
EUVIMED