Multi-Task Mixture-of-Experts Model for Underwater Target Localization and Recognition
Remote Sensing · 26.08.2025 · 17 (17) · 2961
DOI: 10.3390/rs17172961 ↗Kurz gefasst
Fachartikel in der Zeitschrift Remote Sensing, online erschienen am 26.08.2025. Verfasst von Peng Qian und 7 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 71 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Guangdong Basic and Applied Basic Research Foundation.
Abstract (Originalsprache)
The scarcity of underwater acoustic data in deep and remote sea environments poses a significant challenge to data-driven target recognition models, severely restricting their performance. To address this challenge, this study presents a ray-theory-based data augmentation method for generating synthetic ship-radiated noise datasets in oceanic environments at a depth of 3500 m—DS3500, encompassing both direct and shadow zones. Additionally, a novel MEG (multi-task, multi-expert, multi-gate) framework is developed to achieve simultaneous target localization and recognition by integrating relative positional information between the target and sonar, which dynamically partitions parameter spaces through multi-expert mechanisms and adaptively combines task-specific representations using multi-gate attention to simultaneously predict target localization and recognition. Experimental results on the DS3500 dataset demonstrate that the MEG framework achieves 95.93% recognition accuracy, a range localization error of 0.2011 km and a depth localization error of 20.61 m with a maximum detection range of 11 km and depth of 1100 m. This study provides a new technical solution for underwater acoustic target recognition in deep and remote seas, offering innovative approaches for practical applications in marine monitoring and defense.
Förderung
- National Natural Science Foundation of China
- Guangdong Basic and Applied Basic Research Foundation
Zitieren
APA
Qian, P., Wang, J., Liu, Y., Chen, Y., Wang, P., Deng, Y., Xiao, P., & Li, Z. (2025). Multi-Task Mixture-of-Experts Model for Underwater Target Localization and Recognition. Remote Sensing, 17(17), 2961. https://doi.org/10.3390/rs17172961
Vancouver
Qian P, Wang J, Liu Y, Chen Y, Wang P, Deng Y, et al. Multi-Task Mixture-of-Experts Model for Underwater Target Localization and Recognition. Remote Sensing. 2025;17(17):2961. doi:10.3390/rs17172961
BibTeX
@article{Qian2025,
author = {Qian, Peng and Wang, Jingyi and Liu, Yining and Chen, Yingxuan and Wang, Pengjiu and Deng, Yanfa and Xiao, Peng and Li, Zhenglin},
title = {Multi-Task Mixture-of-Experts Model for Underwater Target Localization and Recognition},
journal = {Remote Sensing},
year = {2025},
volume = {17},
number = {17},
pages = {2961},
doi = {10.3390/rs17172961}
}
EUVIMED