EUVIMEDEUVIMED entdecken
Crossref

COMPARATIVE ANALYSIS OF YOLOV8 AND YOLOV10 FOR REAL-TIME OBJECT DETECTION ON A SPECIALIZED MILITARY DATASET

S. I. Hlod, A. V. Doroshenko

Ukrainian Journal of Information Technology · 2026 · 8 (1) · 51-59

DOI: 10.23939/ujit2026.01.051 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Ukrainian Journal of Information Technology, 2026. Verfasst von S. I. Hlod und 1 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid evolution of one-stage object detection algorithms has fundamentally reshaped real-time computer vision, enabling accurate, low-latency perception in dynamic, resource-constrained environments. Among these approaches, the You Only Look Once (YOLO) family has emerged as a dominant framework due to its efficient end-to-end detection pipeline and continuous architectural refinement. This study presents a comprehensive analytical and experimental comparison of two recent representatives of the YOLO family – YOLOv8 and YOLOv10 – focusing on their architectural design principles, performance characteristics, and suitability for real-time deployment. Each version introduces distinct improvements in feature extraction, attention mechanisms, and computational efficiency, thereby enhancing detection accuracy across diverse operational contexts, with particular emphasis on edge and embedded systems. It is established that YOLOv8 introduces an anchor-free detection paradigm, combined with optimized C2 f backbone modules and multi-scale feature fusion, to achieve a balanced trade-off between accuracy and computational efficiency. In contrast, it is determined that YOLOv10 proposes a novel NMS-free detection framework based on a dual-assignment strategy, enabling fully end-to-end inference and reducing post-processing overhead. This work systematically reviews recent literature to highlight the strengths and limitations of both models and to examine their evolution within modern one-stage detectors. To validate the analytical findings, a practical experiment was conducted using a specialized military dataset consisting of armored vehicle imagery, including tanks, BMPs, and BTRs, under varied visual conditions. Both models were trained and evaluated under identical settings, and their performance was assessed using standard metrics, including precision, recall, mean average precision, and inference speed on embedded hardware. The results demonstrate that while YOLOv8 provides slightly higher precision and stable classification behavior, YOLOv10 achieves superior recall, higher mean average precision, and faster inference speed. In particular, YOLOv10 reached mAP@50 of 0.938 and mAP@[50‑95] of 0.664, compared to 0.922 and 0.656 for YOLOv8, respectively. Furthermore, YOLOv10 achieved an average inference speed of 61 FPS on NVIDIA Jetson Orin Nano Super, outperforming YOLOv8, which reached 55 FPS under identical conditions. These findings confirm the ongoing refinement of YOLO architectures toward greater efficiency, adaptability, and practical deployment across complex real-world environments.

Zitieren

APA

Hlod, S. I., & Doroshenko, A. V. (2026). COMPARATIVE ANALYSIS OF YOLOV8 AND YOLOV10 FOR REAL-TIME OBJECT DETECTION ON A SPECIALIZED MILITARY DATASET. Ukrainian Journal of Information Technology, 8(1), 51-59. https://doi.org/10.23939/ujit2026.01.051

Vancouver

Hlod SI, Doroshenko AV. COMPARATIVE ANALYSIS OF YOLOV8 AND YOLOV10 FOR REAL-TIME OBJECT DETECTION ON A SPECIALIZED MILITARY DATASET. Ukrainian Journal of Information Technology. 2026;8(1):51-59. doi:10.23939/ujit2026.01.051

BibTeX
@article{Hlod2026,
  author = {Hlod, S. I. and Doroshenko, A. V.},
  title = {COMPARATIVE ANALYSIS OF YOLOV8 AND YOLOV10 FOR REAL-TIME OBJECT DETECTION ON A SPECIALIZED MILITARY DATASET},
  journal = {Ukrainian Journal of Information Technology},
  year = {2026},
  volume = {8},
  number = {1},
  pages = {51--59},
  doi = {10.23939/ujit2026.01.051}
}
Kostenlos starten →