Long baseline underwater source localization based on deep K-Means++ clustering in complex underwater environments

Long baseline localization relies on trilateration, with the least squares method being utilized to determine the unique position of underwater sources. However, it is highly sensitive to distance measurements from any of the reference beacons to the source. Furthermore, achieving continuous and sta...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	Digital signal processing Jg. 164; S. 105281
Hauptverfasser:	Dai, Yawen, Yang, Lei, Cao, Yifei
Format:	Journal Article
Sprache:	Englisch
Veröffentlicht:	Elsevier Inc 01.09.2025
Schlagworte:	Deep K-Means++ clustering Long baseline Stack denoising autoencoder Underwater source localization Long baseline Underwater source localization Deep K-Means++ clustering Stack denoising autoencoder
ISSN:	1051-2004
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	Long baseline localization relies on trilateration, with the least squares method being utilized to determine the unique position of underwater sources. However, it is highly sensitive to distance measurements from any of the reference beacons to the source. Furthermore, achieving continuous and stable high-precision distance measurements presents a significant challenge in complex marine environments. To address this practical problem, this paper proposes a long baseline underwater source localization method based on deep K-Means++ clustering. A three-layer stack denoising autoencoder (SDA) was utilized to extract the features of the trilateration results. Subsequently, the K-Means++ algorithm was utilized to conduct multi-source location cluster analysis and fine-tuning. The experimental results demonstrate that, compared to the existing Kmeans-based long baseline localization method, the proposed approach demonstrates a modest enhancement in localization accuracy. Concurrently, it has led to an average increase of 7 percentage points in normalized mutual information (NMI), a rise of 8 percentage points in the adjusted Rand index (ARI), and an improvement of 2.5 percentage points in clustering accuracy (ACC). This not only ensures the stability and robustness of accuracy against ocean noise in long baseline mode but also enhances the operational efficiency and versatility of clustering methods applied within this field.
ISSN:	1051-2004
DOI:	10.1016/j.dsp.2025.105281