EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Improved YOLOv8n‐seg for Instance Segmentation of the Sonar Images of Marine Debris

Yongjie Jiang, Haitao Guo, Fenghao Xu, Xiaolong Zhu

IET Image Processing · 2026-01 · 20 (1)

DOI: 10.1049/ipr2.70304 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 29.04.2026. Verfasst von Yongjie Jiang und 3 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In order to efficiently clean up marine debris, it is necessary to obtain the types and locations of marine debris. By using instance segmentation on the sonar images of marine debris, the types and locations of marine debris can be obtained. An improved YOLOv8n‐seg (model), for instance, segmentation of the sonar images of marine debris, is proposed in this article. For this purpose, the backbone (network) of the improved YOLOv8n‐seg is designed by proposing the model C2PC‐Cone (cross‐stage partial bottleneck with two partial convolutions and cone activation function) and replacing the module C2f (cross‐stage partial bottleneck with two convolutions) in the backbone of the YOLOv8n‐seg with the C2PC‐Cone; the neck (network) of the improved YOLOv8n‐seg is designed by replacing the module C2f in the neck of the YOLOv8n‐seg with the module VoV‐GSCSP (variety of view gather‐scatter cross‐stage partial); and the head (network) of the improved YOLOv8n‐seg, named the SA‐segment, is designed by introducing the self‐attention module. Experimental results on the selected dataset of the sonar images of marine debris show that the improved YOLOv8n‐seg achieved P (precision) of 83.6%, R (recall) of 87.9%, mAP@50 (mean average precision at IoU 0.5, where IoU denotes Intersection over Union) of 88.9%, and mAP@50‐95 (mean average precision at IoU range 0.5–0.95) of 54.2%, with only 8.9G FLOPs (floating point operations) and 2.550 M NOPs (number of parameters). For instance segmentation on the selected dataset of the sonar images of marine debris, compared to the YOLOv8n‐seg, the improved YOLOv8n‐seg has better evaluation metrics for P , R , mAP@50, mAP@50‐95, FLOPs, and NOPs. These experimental results validate the effectiveness of the improved YOLOv8n‐seg in enhancing segmentation accuracy while lowering the resource consumption and the computational complexity.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Jiang, Y., Guo, H., Xu, F., & Zhu, X. (2026). Improved YOLOv8n‐seg for Instance Segmentation of the Sonar Images of Marine Debris. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70304

Vancouver

Jiang Y, Guo H, Xu F, Zhu X. Improved YOLOv8n‐seg for Instance Segmentation of the Sonar Images of Marine Debris. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70304

BibTeX
@article{Jiang2026,
  author = {Jiang, Yongjie and Guo, Haitao and Xu, Fenghao and Zhu, Xiaolong},
  title = {Improved YOLOv8n‐seg for Instance Segmentation of the Sonar Images of Marine Debris},
  journal = {IET Image Processing},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/ipr2.70304}
}
Kostenlos starten →