EUVIMEDEUVIMED entdecken
Crossref

A Lightweight Real-Time Sidescan Sonar Image Detection Model Based on a Multiscale Fusion Module

Kun Hao, Lang Xu, Jinjun Liu, Xiaofang Zhao, Zhisheng Li

IEEE Journal of Oceanic Engineering · 2026-01 · 51 (1) · 545-560

DOI: 10.1109/joe.2025.3603271 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Journal of Oceanic Engineering, 2026-01. Verfasst von Kun Hao und 4 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 29 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Ministry of Education.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Sonar image detection plays a crucial role in underwater detection tasks, and accurate sonar image detection technology is an essential guarantee for underwater exploration. To address the issues of high noise, multiple scales, and occlusion in sidescan sonar (SSS) images that lead to a decrease in detection performance, this article proposes a lightweight real-time SSS image detection model, named BAS-YOLOv8. First, by constructing a CSPDarkNet53-CA feature extraction network, the local module is enhanced to suppress background noise, thereby improving the issues of occlusion and high noise in SSS images. Then, to address the problem of multiscale variations of targets in SSS images, we propose a multiscale fusion module BGFpn, which applies a top-down path to the features output by the feature extraction network to guide subsequent network modules for multiscale feature fusion, and uses efficient lightweight convolution (GSConv) to replace ordinary convolution, enhancing the model’s detection accuracy for multiscale SSS image targets. Finally, a channel spatial reconstruction convolution module is introduced in the C2f of the neck module, which solves the problem of reduced feature map resolution with increasing network depth, resulting in a weakened ability to extract small target feature maps and large model parameters. The experiments show that the mAP@0.5 and mAP@0.95 of BAS-YOLOv8 are 93.7% and 65.5% respectively, which are improvements of 3.3% and 2%, respectively, compared to those of YOLOv8n, and the model size is reduced by 10%. Overall, BAS-YOLOv8 features strong robustness, lightweight, fast speed, and high accuracy, meeting the real-time detection requirements of SSS images.

Förderung

  • National Natural Science Foundation of China
  • Ministry of Education

Zitieren

APA

Hao, K., Xu, L., Liu, J., Zhao, X., & Li, Z. (2026). A Lightweight Real-Time Sidescan Sonar Image Detection Model Based on a Multiscale Fusion Module. IEEE Journal of Oceanic Engineering, 51(1), 545-560. https://doi.org/10.1109/joe.2025.3603271

Vancouver

Hao K, Xu L, Liu J, Zhao X, Li Z. A Lightweight Real-Time Sidescan Sonar Image Detection Model Based on a Multiscale Fusion Module. IEEE Journal of Oceanic Engineering. 2026;51(1):545-560. doi:10.1109/joe.2025.3603271

BibTeX
@article{Hao2026,
  author = {Hao, Kun and Xu, Lang and Liu, Jinjun and Zhao, Xiaofang and Li, Zhisheng},
  title = {A Lightweight Real-Time Sidescan Sonar Image Detection Model Based on a Multiscale Fusion Module},
  journal = {IEEE Journal of Oceanic Engineering},
  year = {2026},
  volume = {51},
  number = {1},
  pages = {545--560},
  doi = {10.1109/joe.2025.3603271}
}
Kostenlos starten →