A survey on machine learning in ship radiated noise - passive measurements and future directions

Aug 2, 2026

Upload by Hazel

Machine learning for analyzing ship radiated noise (SR-N) is rapidly evolving, driven by the need to monitor sonar-relevant acoustic levels in highly variable ocean environments. Omnipresent background noise and the scarcity of publicly available labeled datasets introduce major challenges. A comprehensive survey consolidates state-of-the-art SR-N approaches with emphasis on passive measurements, organizing recent work across datasets, data augmentation, signal denoising, feature extraction, detection, localization, and recognition, and outlining promising future research directions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Introduction

Ocean sound pollution and SR-N monitoring

Sonar system types and focus on passive sonar

Passive sonar equation and acoustic components

Survey scope and organization

Datasets and data augmentation

Signal denoising and feature extraction

Detection, localization, and recognition

Future research directions

Ocean Engineering 298 (2024) 117252 | Review
A survey on machine learning in ship radiated noise Hilde I. Hummel a,∗, Rob van der Mei a,b, Sandjai Bhulaib
a Centrum Wiskunde & Informatica, Department of Stochastics, Science Park 123, Amsterdam, 1098 XG, Netherlands b Vrije Universiteit, Department Mathematics, De Boelelaan 1111, Amsterdam, 1081 HV, Netherlands | | | | | --- | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Ship radiated noise Machine learning Survey
Deep learning Underwater sound | | The utilization of machine learning in analyzing ship radiated noise (SR-N) is undergoing rapid evolution. Because the omnipresent background noise strongly depends on the highly variable environment, the application of such techniques poses challenges. Furthermore, publicly available labeled datasets are scarce. Motivated by this, there has been a surge in the number of publications regarding the implementation of machine learning in the monitoring of SR-N within the past few years. This comprehensive survey delineates the stateof-the-art machine learning techniques applied to SR-N, with a specific focus on passive measurements. Recent developments are categorized into several sub-areas, namely; publicly available datasets, data augmentation, signal denoising, feature extraction, detection, localization, and recognition of SR-N. Additionally, future research directions are explored. | | 1. Introduction The health of our ocean environment is endangered by humaninduced sound pollution. The main pollution source originates from noise generated by ships, called Ship Radiated Noise (SR-N). This raises the need to measure and analyze the SR-N levels. This quantifies the amount of pollution and identifies or locates the ships producing noise. The measurement of underwater noise is utilized using so-called sonar systems. These systems can be categorized into three groups: (1) active sonar, (2) side scan sonar, and (3) passive sonar. The active sonar system emits an acoustic pulse and listens to the returning echo. Similarly, the side scan sonar can generate an acoustic image based on the measured echo. This survey will focus on passive sonar systems. These systems do not emit any sound but quietly listen to the noise in the ocean. It passively detects sound waves coming towards the hydrophone(s). This makes the passive system the ultimate system to monitor SR-N. From passive measurements, the SR-N can be analyzed in more detail. This analysis is challenged by multiple factors as expressed in the passive sonar equation: 􀁓􀁎􀁒(dB) = −(􀁎􀁌(dB) − 􀁁􀁇(dB) ) − 􀁔 􀁌(dB) + 􀁓􀁌(dB) . (1) In this equation, all terms have the specific underwater sound unit dB relative to 1 μPa. SNR is the signal-to-noise ratio, which is equivalent to the measured sound by the hydrophone(s). The Noise Level (NL) represents the background noise produced by the ocean environment. The Array Gain (􀁁􀁇) reduces the NL. This value is set to 0 dB for a single hydrophone. Besides the ocean’s background noise, the measurement is ∗ Corresponding author. E-mail address: [h.i.hummel@cwi.nl](h.i.hummel@cwi.nl) (H.I. Hummel). also distorted by Transmission Loss (TL). This is the energy loss of the acoustic source during travel from source to receiver. The total amount of energy loss depends on multiple environmental factors such as water temperature, depth, multi-path distortions, and sea bed type. The TLand NL are highly variable and may change over time since they depend on the dynamic and complex ocean environment. The final element in the sonar equation is the Source Level (SL), which is the acoustic energy level of the SR-N. The acoustic energy level is composed of three elements (Smith and Rigby, 2022; Slamnoiu et al., 2016; Veirs et al., 2016; Urick and United States. Naval Sea Systems Command. Undersea Warfare Technology Office, 1984; Liu et al., 2023c): • machinery noise(20–1000 Hz); • flow noise (5–10 Hz); •

FAQs

Why is analyzing ship radiated noise with machine learning challenging?

Background noise depends strongly on a variable environment, and publicly available labeled datasets are scarce.

Which sonar approach does the survey primarily focus on?

Passive sonar systems, which do not emit sound and instead listen to noise captured by hydrophones.

What areas of SR-N machine learning are covered in the survey?

Techniques are organized into datasets, data augmentation, signal denoising, feature extraction, detection, localization, and recognition, with future research directions discussed as well.

Aug 2, 2026

Upload by Hazel