EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

MSUD-YOLO: A Novel Multiscale Small Object Detection Model for UAV Aerial Images

Xiaofeng Zhao, Hui Zhang, Wenwen Zhang, Junyi Ma, Chenxiao Li, Yao Ding, Zhili Zhang

Drones · 13.06.2025 · 9 (6) · 429

DOI: 10.3390/drones9060429 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Drones, online erschienen am 13.06.2025. Verfasst von Xiaofeng Zhao und 6 weiteren Autor:innen. Die Arbeit wurde bisher 22-mal zitiert (Crossref) und nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, National Foundation for Enhancing Fundamental Sciences in China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Due to the objects in UAV aerial images often presenting characteristics of multiple scales, small objects, complex backgrounds, etc., the performance of object detection using current models is not satisfactory. To address the above issues, this paper designs a multiscale small object detection model for UAV aerial images, namely MSUD-YOLO, based on YOLOv10s. First, the model uses an attention scale sequence fusion mode to achieve more efficient multiscale feature fusion. Meanwhile, a tiny prediction head is incorporated to make the model focus on the low-level features, thus improving its ability to detect small objects. Secondly, a novel feature extraction module named CFormerCGLU has been designed, which improves feature extraction capability in a lighter way. In addition, the model uses lightweight convolution instead of standard convolution to reduce the model’s computation. Finally, the WIoU v3 loss function is used to make the model more focused on low-quality examples, thereby improving the model’s object localization ability. Experimental results on the VisDrone2019 dataset show that MSUD-YOLO improves mAP50 by 8.5% compared with YOLOv10s. Concurrently, the overall model reduces parameters by 6.3%, verifying the model’s effectiveness for object detection in UAV aerial images in complex environments. Furthermore, compared with multiple latest UAV object detection algorithms, our designed MSUD-YOLO offers higher detection accuracy and lower computational cost; e.g., mAP50 reaches 43.4%, but parameters are only 6.766 M.

Förderung

  • National Natural Science Foundation of China
  • National Foundation for Enhancing Fundamental Sciences in China

Zitieren

APA

Zhao, X., Zhang, H., Zhang, W., Ma, J., Li, C., Ding, Y., & Zhang, Z. (2025). MSUD-YOLO: A Novel Multiscale Small Object Detection Model for UAV Aerial Images. Drones, 9(6), 429. https://doi.org/10.3390/drones9060429

Vancouver

Zhao X, Zhang H, Zhang W, Ma J, Li C, Ding Y, et al. MSUD-YOLO: A Novel Multiscale Small Object Detection Model for UAV Aerial Images. Drones. 2025;9(6):429. doi:10.3390/drones9060429

BibTeX
@article{Zhao2025,
  author = {Zhao, Xiaofeng and Zhang, Hui and Zhang, Wenwen and Ma, Junyi and Li, Chenxiao and Ding, Yao and Zhang, Zhili},
  title = {MSUD-YOLO: A Novel Multiscale Small Object Detection Model for UAV Aerial Images},
  journal = {Drones},
  year = {2025},
  volume = {9},
  number = {6},
  pages = {429},
  doi = {10.3390/drones9060429}
}
Kostenlos starten →