EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Lightweight Optimization of YOLO Models for Resource-Constrained Devices: A Comprehensive Review

Ula Kh. Altaie, A.E. Abdelkareem, Abdullah Alhasanat

Diyala Journal of Engineering Sciences · 10.12.2025 · 1-26

DOI: 10.24237/djes.2025.18401 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Diyala Journal of Engineering Sciences, online erschienen am 10.12.2025. Verfasst von Ula Kh. Altaie und 2 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The growing adoption of Internet of Things (IoT) and edge computing has increased the need for effective real-time object detectors that can operate in resource-constrained environments. YOLO (You Only Look Once), a leading state-of-the-art object detection algorithm, is well known for its remarkable real-time performance in a variety of applications. However, traditional YOLO models remain computationally heavy for resource-constrained environments since they are designed for high-performance systems, making them less practical for low-power, embedded platforms such as Raspberry Pi, ARM-based processors, and NVIDIA Jetson edge devices. This paper aims to investigate and analyse optimization strategies that enhance YOLO’s efficiency for edge deployments and provides a comprehensive review of various optimization techniques to overcome the deployment challenges. Two main approaches are explored, structural modification using lightweight modules like ShuffleNet, MobileNet, and GhostNet, and model compression via knowledge distillation, quantization, and pruning. The reviewed works demonstrate significant reductions in model size and complexity, with generally enhanced inference speed and improving accuracy, however, in certain cases, a slight drop in accuracy and frame rate occurs as the cost of achieving higher efficiency. Structural modifications generally support model stability, efficiency, and generalization, while compression-based techniques further improve models’ compactness and inference throughput. A combined or hybrid optimization strategy offers the most balanced solution, achieving strong detection accuracy with reductions in model size, GFLOPs, and overall inference cost. This narrative synthesis review provides guidance for developing scalable, energy-aware YOLO models suitable for edge-based detection applications in fields like autonomous vehicles, smart cities, and IoT-driven systems.

Zitieren

APA

Altaie, U. K., Abdelkareem, A., & Alhasanat, A. (2025). Lightweight Optimization of YOLO Models for Resource-Constrained Devices: A Comprehensive Review. Diyala Journal of Engineering Sciences, 1-26. https://doi.org/10.24237/djes.2025.18401

Vancouver

Altaie UK, Abdelkareem A, Alhasanat A. Lightweight Optimization of YOLO Models for Resource-Constrained Devices: A Comprehensive Review. Diyala Journal of Engineering Sciences. 2025:1-26. doi:10.24237/djes.2025.18401

BibTeX
@article{Altaie2025,
  author = {Altaie, Ula Kh. and Abdelkareem, A.E. and Alhasanat, Abdullah},
  title = {Lightweight Optimization of YOLO Models for Resource-Constrained Devices: A Comprehensive Review},
  journal = {Diyala Journal of Engineering Sciences},
  year = {2025},
  pages = {1--26},
  doi = {10.24237/djes.2025.18401}
}
Kostenlos starten →