Vessel Length Estimation from Magnetic Wake Signature: A Physics-Informed Residual Neural Network Approach
Mohammad Amir Fallah, Mehdi Monemi, Member, IEEE, Matti Latva-aho, Fellow, IEEE · 2025
Read Vessel Length Estimation from Magnetic Wake Signature: A Physics-Informed Residual Neural Network Approach by Mohammad Amir Fallah, Mehdi Monemi, Member, IEEE, Matti Latva-aho, Fellow, IEEE (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-19112.txt.
Introduction
Mohammad Amir Fallah and co-authors estimate a ship's length from the faint magnetic wake it leaves behind. Knowing vessel length aids classification and surveillance at sea. Older methods read the water wake in images or demand costly satellite pictures; a single airborne magnetic scan is far cheaper.
The authors first write equations linking the magnetic wake to vessel length in water of realistic depth. A deep residual neural network, guided by that physics, then solves them from a one-dimensional scan profile. Physics-informed learning keeps the network honest where data are thin. The approach could sharpen maritime monitoring from aircraft at modest cost.
Catalog record
| CREATOR | Mohammad Amir Fallah, Mehdi Monemi, Member, IEEE, Matti Latva-aho, Fellow, IEEE |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-19112
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Vessel Length Estimation from Magnetic Wake Signature: A Physics-Informed Residual Neural Network Approach Mohammad Amir Fallah, Mehdi Monemi, Member, IEEE, Matti Latva-aho, Fellow, IEEE Abstract Marine remote sensing enhances maritime surveillance, environmental monitoring, and naval operations. Vessel length estimation, a key component of this technology, supports effective maritime surveillance by empowering features such as vessel classification. Departing from traditional methods relying on two-dimensional hydrodynamic wakes or computationally intensive satellite imagery, this paper introduces an innovative approach for vessel length estimation that leverages the subtle magnetic wake signatures of vessels, captured through a low-complexity one-dimensional profile from a single airborne magnetic sensor scan. The proposed method centers around our characterized nonlinear integral equations that connect the magnetic wake to the vessel length within a realistic finite-depth marine environment. To solve the derived equations, we initially leverage a deep residual neural network (DRNN). The proposed DRNN-based solution framework is shown to be unable to exactly learn the intricate relationships between parameters when constrained by a limited training-dataset. To overcome this issue, we introduce an innovative approach leveraging a physics-informed residual neural network (PIRNN). This model integrates physical formulations directly into the loss function, leading to improved performance in terms of both accuracy and convergence speed. Considering a sensor scan angle of less than 15 ∘ 15^{\circ} , which maintains a reasonable margin below Kelvin’s limit angle of 19.5 ∘ 19.5^{\circ} , we explore the impact of various parameters on the accuracy of the vessel length estimation, including sensor scan angle, vessel speed, and sea depth. Numerical simulations demonstrate the superiority of the proposed PIRNN method, achieving mean length estimation errors consistently below 5% for vessels longer than 100m. For shorter vessels, the errors generally remain under 10%. I Introduction Maritime surveillance and vessel traffic control have consistently been of critical importance, propelling the advancement of state-of-the-art remote sensing systems tailored for marine environments. The emergence of technologically advanced marine vessels has highlighted the need to modernize existing detection technologies, including those based on Synthetic Aperture Radar (SAR), lidar, and sonar methodologies. The principal objective of these technologies remains vessel identification through the detection of anomalies, employing two primary approaches: active and passive detection. Active detection methods involve transmitting a signal and analyzing its reflection from the target, while passive methods rely on distinguishing the vessel’s signal from the background environmental noise. Active detection methods, including Lidar and SAR, while capable of providing precise target localization, are vulnerable to countermeasures such as jamming and deception, thereby compromising their operational reliability [ 1 ] . Moreover, these active airborne techniques are typically limited to the detection of surface vessels and are ineffective for detecting submerged targets. Consequently, contemporary research increasingly focuses on the development of passive methods for detecting distant vessels. Sonar-based systems have been a popular choice due to their low acoustic path loss in seawater. However, challenges such as interference from unwanted acoustic signals, intentional deception techniques, and advances in stealth vessel technology have limited their effectiveness. To address these limitations, researchers have increasingly turned to magnetic anomaly detection (MAD) as a promising alternative [ 2 , 3 ] . MAD leverages the magnetic field disturbances generated by moving vessels to identify their presence. While traditional MAD methods focus on detecting vessels with metallic hulls, recent advancements in degaussing technologies have necessitated the exploration of new approaches. One such approach involves detecting the hydrodynamic anomalies induced by moving vessels [ 4 , 5 ] . This method focuses on identifying hydrodynamic anomalies known as hydrodynamic wakes [ 6 ] . The hydrodynamic wake has two regimes of near and far field; the later called as Kelvin waves is involved in the detection of the remote vessel [ 7 , 8 , 9 ] . The movement of water as a weak conductor in the natural magnetic field of the earth induces magnetic anomaly whose behavior is very similar to the original hydrodynamic Kelvin waves and thus can lead to the detection of the remote vessel if precisely measured and processed with a magnetometer [ 10 , 11 ] . A key characteristic of magnetic anomaly induced by the moving vessel is that it lasts for a long time and can be sensed at large distances which makes it a good tool as a passive remote detection method in sea [ 12 , 7 , 11 , 13 ] . On the other hand, the aforementioned magnetic anomaly does not depend on the vessel body material, and provides the feasibility of detecting remote vessels equipped with degaussing tools or having non-metallic hulls. From a remote sensing perspective, beyond mere vessel detection, the acquisition of static and dynamic information, such as speed, heading, and hull dimensions, is also of considerable interest. The estimation of vessel speed and heading using magnetic wake analysis has been addressed in prior research [ 12 , 7 , 5 , 4 ] . However, the estimation of vessel length from hydrodynamic wakes has been a subject of ongoing investigation, with various techniques proposed for extracting this information from remote sensing data. Early studies, such as that of [ 14 ] , focused on analyzing the spectral characteristics of 2D hydrodynamic vessel wakes. By examining parameters like peak frequency, researchers were able to develop theoretical models for vessel length estimation. While these approaches show potential, their accuracy is sensitive to environmental factors, including assumptions of infinite water depth, wind conditions, and sea state, all of which can substantially influence their precision. More recent research has leveraged advanced image processing techniques to directly estimate vessel dimensions from satellite imagery. Authors in [ 15 ] introduced a method for accurately outlining vessels in Sentinel-1 SAR images. This method, which employed image processing techniques to refine vessel shapes, allowed for precise measurements of vessel length and width. While this approach demonstrated high accuracy, particularly in estimating vessel length, its performance was hindered by limitations in image resolution and the presence of significant sea clutter. The application of deep learning has significantly advanced the field of vessel size estimation. A deep learning model, known as SSENet , has been specifically designed to directly estimate the size of vessels from Sentinel-1 Synthetic Aperture Radar images [ 16 ] . While demonstrating impressive performance, the model’s dependency on large-scale accurately labeled datasets as well as the high computational complexity of the solution scheme limit its practical application. Beyond size estimation, SAR imagery has also been explored to extract vessel hull characteristics from their wake patterns [ 17 ] . The use of features like the width, length, and intensity of the wake to classify vessel types and estimate their size and speed is investigated. While SAR imagery offers unique advantages, optical imagery remains a valuable resource for vessel size estimation. A method has been presented to estimate vessel size from satellite optical images by fitting an ellipse to the vessel’s outline [ 18 ] . In this work, we focus predominantly on estimating the length of a vessel through the analysis of its magnetic wake. Unlike traditional methods that rely on analyzing the 2D hydrodynamic wake , this research exclusively utilizes a one-dimensional data acquisition of the magnetic wake ( 1D magnetic wake ) to determine