pith. sign in

arxiv: 2606.21326 · v1 · pith:7ZP2D2NWnew · submitted 2026-06-19 · 💻 cs.SD · cs.LG· eess.SP

Sea-Scan: High-Accuracy, ML-based Dark Vessel Detection and Localisation via Weakly Supervised DAS Monitoring

Tian Tian , Agastya Raj , Lara Flanagan , John Kennedy , Marco Ruffini This is my paper

Pith reviewed 2026-06-26 13:04 UTC · model grok-4.3

classification 💻 cs.SD cs.LGeess.SP
keywords dark vessel detectionweakly supervised learningDAS monitoringvessel localizationmachine learningmaritime surveillanceacoustic sensing
0
0 comments X

The pith

A weakly supervised ML model detects dark vessels at 97.8% rate and localizes them in DAS data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an ML system that detects and localizes vessels not reporting via AIS, using weak supervision from imperfect AIS labels on Distributed Acoustic Sensing recordings. The approach reaches 97.8% detection at 1.98% false-trigger rate and flags dark-vessel events in data lacking any labels. A reader would care because dark vessels evade conventional tracking systems, and the method avoids the cost of dense manual annotations for training. The work focuses on practical maritime monitoring where cooperative signals are absent or unreliable.

Core claim

We present an ML-based vessel detection and localization system, trained with weak supervision from imperfect AIS labels, that achieves a 97.8% detection rate at 1.98% false-trigger rate, successfully identifies dark-vessel events from unlabeled data.

What carries the argument

Weakly supervised ML model trained on DAS acoustic data with AIS labels for vessel detection and localization

If this is right

  • The system detects vessels absent from AIS records.
  • It provides both detection and localization outputs.
  • Performance holds on completely unlabeled DAS data.
  • Weak supervision removes the need for precise per-event labels during training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the model generalizes across sea states and sensor deployments, it could support large-scale passive monitoring of exclusive economic zones.
  • The same weak-supervision strategy might transfer to other acoustic or optical maritime sensors without new labeling campaigns.
  • Cross-validation against satellite or radar dark-vessel catalogs would provide an external check on whether the reported rates hold in operational settings.

Load-bearing premise

Imperfect AIS labels supply enough reliable signal for the model to learn dark-vessel patterns that transfer to new unlabeled DAS recordings.

What would settle it

Running the trained model on a separate DAS dataset containing independently confirmed dark vessel positions and observing detection rates well below 97.8% or false-trigger rates well above 1.98%.

Figures

Figures reproduced from arXiv: 2606.21326 by Agastya Raj, John Kennedy, Lara Flanagan, Marco Ruffini, Tian Tian.

Figure 1
Figure 1. Figure 1: End-to-end pipeline of the proposed DAS vessel detection and localization framework. a binary corridor mask: channels within ±2 km are labeled positive. This radius accounts for un￾certainty from projection error and lateral extent of seabed acoustic coupling. The mask is inten￾tionally conservative, providing candidate-positive regions that the model must learn to refine. Model Structure. The encoder back… view at source ↗
Figure 2
Figure 2. Figure 2: Detection example across 120 km of cable over a 2-hour interval. (a) EFBL cable route and AIS vessel trajectories. (b) Multi-band DAS envelope intensity. (c) AIS-reported vessel characteristics for the four crossings. (d) AIS-derived mask of proximity corridor, projected on to the time-distance grid. (e) Temporal activity score Pt for each 5 km cable segment. (f) Final confidence map Mt,c with threshold co… view at source ↗
read the original abstract

We present an ML-based vessel detection and localization system, trained with weak supervision from imperfect AIS labels, that achieves a 97.8% detection rate at 1.98% false-trigger rate, successfully identifies dark-vessel events from unlabeled data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents Sea-Scan, an ML-based vessel detection and localization system trained via weak supervision from imperfect AIS labels on DAS data. It reports a 97.8% detection rate at 1.98% false-trigger rate and claims that the model successfully identifies dark-vessel events from unlabeled data.

Significance. If the generalization to truly unlabeled dark-vessel detection were quantitatively validated, the approach could offer a practical advance in maritime monitoring by leveraging existing imperfect AIS for training while extending to non-AIS vessels. The current evidence, however, does not yet establish this step.

major comments (2)
  1. [Abstract] Abstract: the headline performance numbers (97.8% detection rate, 1.98% FTR) are presented without any information on dataset size, train/test partitioning, validation procedure, baseline methods, or handling of label noise from imperfect AIS. This omission makes it impossible to determine whether the metrics support the stated claims.
  2. [Results] Results section (inferred from abstract claims): the assertion that the system 'successfully identifies dark-vessel events from unlabeled data' rests on an unquantified generalization step. Because dark vessels are defined by the absence of AIS, any numeric success rate must be supported by an independent verification method (expert review, synthetic dark-vessel injection, or cross-sensor confirmation) on unlabeled segments; the manuscript appears to report metrics exclusively on AIS-matched folds, leaving the central dark-vessel claim without direct quantitative backing.
minor comments (1)
  1. [Abstract] Abstract: a single sentence on the scale or characteristics of the DAS recordings would help readers contextualize the reported rates.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review of our manuscript. We address each major comment below and describe the planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline performance numbers (97.8% detection rate, 1.98% FTR) are presented without any information on dataset size, train/test partitioning, validation procedure, baseline methods, or handling of label noise from imperfect AIS. This omission makes it impossible to determine whether the metrics support the stated claims.

    Authors: We agree that the abstract would benefit from additional experimental context. In the revised version we will expand the abstract to report dataset size, train/test partitioning, validation procedure, baseline comparisons, and the treatment of AIS label noise within the weak-supervision pipeline. revision: yes

  2. Referee: [Results] Results section (inferred from abstract claims): the assertion that the system 'successfully identifies dark-vessel events from unlabeled data' rests on an unquantified generalization step. Because dark vessels are defined by the absence of AIS, any numeric success rate must be supported by an independent verification method (expert review, synthetic dark-vessel injection, or cross-sensor confirmation) on unlabeled segments; the manuscript appears to report metrics exclusively on AIS-matched folds, leaving the central dark-vessel claim without direct quantitative backing.

    Authors: The referee correctly notes that quantitative metrics are computed on AIS-matched folds where ground truth is available. The manuscript currently supports the dark-vessel claim with qualitative examples on unlabeled segments. To strengthen the central claim we will add an independent quantitative validation step (synthetic injection or cross-sensor confirmation) on unlabeled data in the revised results section. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical ML claims self-contained

full rationale

The provided abstract and context contain no equations, derivations, self-citations, or load-bearing steps matching any enumerated circularity pattern. The reported metrics (97.8% detection, 1.98% FTR) are standard held-out performance figures from a weakly supervised model; they do not reduce by construction to the AIS labels via self-definition or renaming. No uniqueness theorems, ansatzes, or fitted inputs presented as independent predictions appear. The paper's central claim rests on empirical generalization rather than any definitional equivalence to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no model details, hyperparameters, or assumptions stated. Cannot enumerate free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5575 in / 908 out tokens · 16918 ms · 2026-06-26T13:04:48.201445+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

11 extracted references · 3 canonical work pages · 2 internal anchors

  1. [1]

    A novel anomaly detection ap- proach to identify intentional ais on-off switching,

    F . Mazzarella, M. Vespe, A. Alessandrini, D. Tarchi, G. Aulicino, and A. Vollero, “A novel anomaly detection ap- proach to identify intentional ais on-off switching,”Expert Systems with Applications,vol. 78, 02 2017

    2017

  2. [2]

    Ocean space surveillance and real-time event characterization using distributed acoustic sensing on submarine networks,

    A. F . Bairdet al., “Ocean space surveillance and real-time event characterization using distributed acoustic sensing on submarine networks,”Seismol. Res. Lett., 96(2A), 691– 705, Jan. 2025

    2025

  3. [3]

    Overview of distributed acoustic sensing: Theory and ocean applications,

    A. Xenaki, P . Gerstoft, E. Williams, and S. Abadi, “Overview of distributed acoustic sensing: Theory and ocean applications,”J. Acoust. Soc. Am., 158(1), 801– 825, Jul. 2025

    2025

  4. [4]

    Preliminary assessment of ship detection and trajectory evaluation using distributed acoustic sens- ing on an optical fiber telecom cable,

    D. Rivetet al., “Preliminary assessment of ship detection and trajectory evaluation using distributed acoustic sens- ing on an optical fiber telecom cable,”J. Acoust. Soc. Am., 149(4), 2615–2627, Apr. 2021

    2021

  5. [5]

    Sensing whales, storms, ships and earthquakes using an Arctic fibre optic cable,

    M. Landrøet al., “Sensing whales, storms, ships and earthquakes using an Arctic fibre optic cable,”Sci. Rep., 12(1), 19226, Nov. 2022

    2022

  6. [6]

    Leveraging distributed acoustic sensing for monitoring vessels using submarine fiber-optic cables,

    B. Paapet al., “Leveraging distributed acoustic sensing for monitoring vessels using submarine fiber-optic cables,” Appl. Ocean Res., 154, 104422, Jan. 2025

    2025

  7. [7]

    DAShip: A large-scale annotated dataset for ship detection using distributed acoustic sensing tech- nique,

    W. Huanget al., “DAShip: A large-scale annotated dataset for ship detection using distributed acoustic sensing tech- nique,”IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., 18, 4093–4107, 2025

    2025

  8. [8]

    Vessel Detection and Localization Using Distributed Acoustic Sensing in Submarine Optical Fiber Cables

    E. E. Ramirez-Torreset al., “Vessel detection and local- ization using distributed acoustic sensing in submarine optical fiber cables,” arXiv:2509.11614, Sep. 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    A deep learning framework for marine acoustic and seismic monitoring with distributed acoustic sensing,

    C. Zhang, W. Zhu, B. A. Romanowicz, R. M. Allen, K. Soga, and Y . Wu, “A deep learning framework for marine acoustic and seismic monitoring with distributed acoustic sensing,” arXiv:2603.14844, Mar. 2026

    work page internal anchor Pith review arXiv 2026

  10. [10]

    Tracking moving ships using distributed acoustic sensing data,

    J. Shao, Y . Wang, Y . Zhang, X. Zhang, and C. Zhang, “Tracking moving ships using distributed acoustic sensing data,”IEEE Geosci. Remote Sens. Lett., 22, 1–5, 2025

    2025

  11. [11]

    UniFormer: Unified transformer for efficient spatiotemporal representation learning,

    K. Liet al., “UniFormer: Unified transformer for efficient spatiotemporal representation learning,” arXiv:2201.04676, 2022

    work page arXiv 2022