Deep-learning-based underwater target recognition: A multi-cycle range–bearing image fusion approach
The Journal of the Acoustical Society of America · 01.11.2026 · 160 (5) · 4335-4343
DOI: 10.1121/10.0046852 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Journal of the Acoustical Society of America, online erschienen am 19.11.2026. Verfasst von Tianfeng Huang und 5 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Shanghai Pujiang Programme.
Abstract (Originalsprache)
We propose a multi-cycle range–bearing image-fusion-based deep-learning framework for underwater target recognition. Within this framework, two multi-cycle image construction strategies are developed: multi-cycle temporal accumulation (MCTA) and multi-cycle spatial concatenation (MCSC), which jointly encode traditional target characteristics, including intensity, spatial scale, and motion-related information. Their recognition performance was systematically evaluated under varying signal-to-noise ratio (SNR) conditions. Results show that both MCTA and MCSC achieve competitive recognition performance compared with single-cycle data under high-SNR conditions. However, under low-SNR conditions, MCSC outperforms MCTA by preserving target structural variations and suppressing background clutter. Recognition accuracy increases with the number of fused cycles and reaches 94.55% when 20 cycles are used. Model interpretability was further investigated using gradient-weighted class activation mapping (Grad-CAM) and t-distributed stochastic neighbor embedding (t-SNE) analyses. t-SNE visualization results show that multi-cycle range–bearing image fusion produces feature representations with larger inter-class separability and more compact intra-category distributions than those obtained from single-cycle images. Grad-CAM results indicate that compared with single-cycle images, 20-cycle MCSC images guide the model to focus more consistently on target-related regions while suppressing responses to background clutter and environmental interference. These findings demonstrate that spatial concatenation effectively aggregates multi-cycle target information, providing a robust and interpretable approach for underwater target recognition under different SNR conditions.
Förderung
- National Natural Science Foundation of China
- Shanghai Pujiang Programme
Zitieren
APA
Huang, T., Wu, K., Ye, T., Yao, X., Li, M., & Guo, M. (2026). Deep-learning-based underwater target recognition: A multi-cycle range–bearing image fusion approach. The Journal of the Acoustical Society of America, 160(5), 4335-4343. https://doi.org/10.1121/10.0046852
Vancouver
Huang T, Wu K, Ye T, Yao X, Li M, Guo M. Deep-learning-based underwater target recognition: A multi-cycle range–bearing image fusion approach. The Journal of the Acoustical Society of America. 2026;160(5):4335-4343. doi:10.1121/10.0046852
BibTeX
@article{Huang2026,
author = {Huang, Tianfeng and Wu, Kun and Ye, Tianming and Yao, Xin and Li, Mei and Guo, Miao},
title = {Deep-learning-based underwater target recognition: A multi-cycle range–bearing image fusion approach},
journal = {The Journal of the Acoustical Society of America},
year = {2026},
volume = {160},
number = {5},
pages = {4335--4343},
doi = {10.1121/10.0046852}
}
EUVIMED