pith. sign in

arxiv: 2606.10811 · v1 · pith:O2VBO2SGnew · submitted 2026-06-09 · 💻 cs.CV

Deep learning for echo sounder data

Ketil Malde This is my paper

Pith reviewed 2026-06-27 13:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords deep learningecho sounderacoustic dataechogramsunderwater acousticsmachine learningdata formatsdatasets
0
0 comments X

The pith

Acoustic echo sounder data requires deep learning methods designed for its intrinsic properties rather than adapted from image processing.

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

The paper argues that deep learning has transformed image and text analysis but produced only modest results when applied to underwater acoustic observations from echo sounders. It claims that the distinctive character of acoustic data means further progress depends on creating new methods instead of reusing image-based models. The main barrier today is the absence of consistent data formats and high-quality public datasets that include agreed performance targets. Removing these barriers would allow targeted research that could improve how acoustic data is interpreted for underwater monitoring.

Core claim

The central claim is that due to intrinsic properties of acoustic data, substantial advances will likely require research into deep learning methods beyond mere recycling of models and techniques from image processing. Currently, the potential for breakthroughs in method development is hindered by the lack of standard data formats and organization, and even more by the lack of readily available, high quality data sets with established performance goals. To advance the field, these shortcomings should be remedied.

What carries the argument

The intrinsic properties of acoustic data that set it apart from images and require purpose-built deep learning approaches for echograms and related recordings.

If this is right

  • New deep learning architectures will need to be developed specifically for the structure of acoustic recordings rather than transferred from vision tasks.
  • Standard data formats and consistent organization practices must be adopted across the field.
  • Public high-quality datasets accompanied by defined performance targets will become necessary to measure and drive progress.
  • Underwater observation tasks that rely on echo sounder data will see improved processing once these resources exist.

Where Pith is reading between the lines

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

  • Marine monitoring applications such as fish detection or seabed mapping could gain from methods that respect acoustic signal characteristics like range-dependent resolution.
  • Similar data-format and benchmark gaps may exist in other sensor domains that produce non-image time-series data.
  • Community efforts to curate and share acoustic recordings would likely accelerate method testing even before new model families appear.

Load-bearing premise

The lack of standard data formats, organization, and readily available high-quality data sets with established performance goals is the primary current hindrance to breakthroughs.

What would settle it

A public release of standardized high-quality acoustic datasets with clear benchmarks followed by rapid development of image-processing-derived models that match or exceed specialized methods on those benchmarks.

read the original abstract

There is no doubt that over the last decade, techniques from the field of machine learning have revolutionized how we process and interpret data, especially images and text. For underwater observations acoustics is a primary source of information, and naturally, deep learning methods have been applied to echograms and other acoustics data, but so far with rather modest results. Here, we argue that due to intrinsic properties of acoustic data, substantial advances will likely require research into deep learning methods beyond mere recycling of models and techniques from image processing. Currently, the potential for breakthroughs in method development is hindered by the lack of standard data formats and organization, and even more by the lack of readily available, high quality data sets with established performance goals. To advance the field, these shortcomings should be remedied

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 / 0 minor

Summary. The manuscript argues that deep learning has produced only modest results when applied to underwater acoustic data such as echograms, attributes this to intrinsic properties of acoustic data that make standard image-processing models insufficient, and identifies the lack of standardized data formats and high-quality public datasets with established benchmarks as the primary barriers to progress; it concludes by calling for remediation of these data shortcomings.

Significance. If the central claim were supported by concrete examples of acoustic-specific failure modes or successful tailored architectures, the perspective could usefully direct community effort toward data curation and domain-adapted methods rather than direct transfer of vision models. In its current form the argument remains an unsupported assertion, limiting its ability to shape research priorities.

major comments (2)
  1. [Abstract] Abstract: the claim that 'modest results' stem from 'intrinsic properties of acoustic data' is presented without enumerating any such properties (e.g., range-dependent spreading, frequency-dependent attenuation, multi-beam geometry, or non-stationary reverberation) or citing even one documented case in which an image-derived CNN or transformer demonstrably underperforms relative to a purpose-built acoustic model.
  2. [Abstract] Abstract: no quantitative evidence, performance tables, or literature citations are supplied to establish the baseline of 'rather modest results' against which future advances would be measured, rendering the necessity of 'research into deep learning methods beyond mere recycling' an assertion rather than a demonstrated conclusion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. As authors of this perspective piece, we appreciate the opportunity to clarify our arguments and will revise the manuscript to better support the claims while preserving its concise, forward-looking nature.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'modest results' stem from 'intrinsic properties of acoustic data' is presented without enumerating any such properties (e.g., range-dependent spreading, frequency-dependent attenuation, multi-beam geometry, or non-stationary reverberation) or citing even one documented case in which an image-derived CNN or transformer demonstrably underperforms relative to a purpose-built acoustic model.

    Authors: The manuscript is a short perspective intended to stimulate discussion rather than serve as an empirical study or exhaustive review. The listed acoustic properties are standard in the underwater acoustics literature and explain why direct transfer of image models is often insufficient. We will revise the abstract to briefly enumerate key properties and add 1-2 targeted citations to published cases demonstrating underperformance of standard CNNs/transformers on echogram data relative to acoustic-aware approaches. revision: partial

  2. Referee: [Abstract] Abstract: no quantitative evidence, performance tables, or literature citations are supplied to establish the baseline of 'rather modest results' against which future advances would be measured, rendering the necessity of 'research into deep learning methods beyond mere recycling' an assertion rather than a demonstrated conclusion.

    Authors: We agree that referencing specific performance levels from the literature would strengthen the perspective. The observation of modest results reflects the authors' synthesis of existing applications (e.g., fish classification and seabed mapping tasks). In revision we will incorporate citations to representative studies along with their reported metrics to establish a clearer baseline without turning the piece into a review article. revision: partial

Circularity Check

0 steps flagged

Position paper with no derivation chain or quantitative claims; central assertion is non-circular by absence of reduction

full rationale

The manuscript is an opinion piece arguing that acoustic data properties require DL methods beyond image processing and that data standardization hinders progress. No equations, fitted parameters, predictions, self-citations, or uniqueness theorems appear in the abstract or described content. The strongest claim is a forward-looking assertion without any derivation that reduces to its inputs by construction, matching the reader's assessment of score 0.0. No load-bearing steps exist to analyze under the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, fitted parameters, or new entities are introduced; the document rests on the domain assumption that acoustic data possess intrinsic properties that make image models unsuitable, but provides no further breakdown.

pith-pipeline@v0.9.1-grok · 5644 in / 1071 out tokens · 18549 ms · 2026-06-27T13:16:49.410769+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

28 extracted references · 2 linked inside Pith

  1. [1]

    Bahlburg, Dominik, Sebastian Menze, Bjørn A Krafft, Andy D Lowther, and Bettina Meyer. 2025. ``Mapping Encounters Between Antarctic Krill Fishing Vessels and Air-Breathing Krill Predators Using Acoustic Data from the Fishery.'' Proceedings of the National Academy of Sciences 122 (25): e2417203122

    2025

  2. [2]

    Bokhovkin, Alexey, and Evgeny Burnaev. 2019. ``Boundary Loss for Remote Sensing Imagery Semantic Segmentation.'' In International Symposium on Neural Networks, 388--401. Springer

    2019

  3. [3]

    Malde, R

    Brautaset, O., K. Malde, R. J. Korneliussen, and N. O. Handegard. 2020. ``Acoustic Fish Detection Using Deep Learning.'' ICES Journal of Marine Science 77 (10): 3614--27

    2020

  4. [4]

    Coetzee, Janet. 2000. ``Use of a Shoal Analysis and Patch Estimation System (SHAPES) to Characterise Sardine Schools.'' Aquatic Living Resources 13 (1): 1--10

    2000

  5. [5]

    Deng, Jia, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. ``Imagenet: A Large-Scale Hierarchical Image Database.'' In 2009 IEEE Conference on Computer Vision and Pattern Recognition, 248--55. IEEE

    2009

  6. [6]

    Deng, Li. 2012. ``The Mnist Database of Handwritten Digit Images for Machine Learning Research [ Best of the Web ] .'' IEEE Signal Processing Magazine 29 (6): 141--42

    2012

  7. [7]

    Dosovitskiy, Alexey, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, et al. 2020. ``An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale.'' arXiv Preprint arXiv:2010.11929

    Pith/arXiv arXiv 2020

  8. [8]

    Echoview Software Pty Ltd. 2025. Echoview Version 15.1. https://www.echoview.com

    2025

  9. [9]

    Everingham, Mark, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. 2010. ``The Pascal Visual Object Classes (Voc) Challenge.'' International Journal of Computer Vision 88 (2): 303--38

    2010

  10. [10]

    Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. 2016. Deep Learning. MIT Press

    2016

  11. [11]

    He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. ``Deep Residual Learning for Image Recognition.'' In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 770--78

    2016

  12. [12]

    Hirama, Yudai, Soichiro Yokoyama, Tomohisa Yamashita, Hidenori Kawamura, Keiji Suzuki, and Masaaki Wada. 2017. ``Discriminating Fish Species by an Echo Sounder in a Set-Net Using a CNN.'' In 2017 21st Asia Pacific Symposium on Intelligent and Evolutionary Systems (IES), 112--15. IEEE

    2017

  13. [13]

    Korneliussen, Rolf J, Yngve Heggelund, Gavin J Macaulay, Daniel Patel, Espen Johnsen, and Inge K Eliassen. 2016. ``Acoustic Identification of Marine Species Using a Feature Library.'' Methods in Oceanography 17: 187--205

    2016

  14. [14]

    Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E Hinton. 2012. ``Imagenet Classification with Deep Convolutional Neural Networks.'' Advances in Neural Information Processing Systems 25

    2012

  15. [15]

    Li, Yangdong, Qinghong Mao, Zhuang Chen, and Guoping Zhu. 2025. ``Enhancing Multi-Frequency Acoustic Signal Extraction of Antarctic Krill Euphausia Superba Using u-Net Convolutional Neural Network.'' Marine Ecology Progress Series 760: 55--69

    2025

  16. [16]

    Lin, Tsung-Yi, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. ``Microsoft Coco: Common Objects in Context.'' In European Conference on Computer Vision, 740--55. Springer

    2014

  17. [17]

    Macaulay, Gavin, and Hector Peña. 2018. The SONAR-netCDF4 Convention for Sonar Data, Version 1.0. ICES Cooperative Research Reports (CRR)

    2018

  18. [18]

    Marques, Tiago A. et al. 2020. ``Detection and Classification of Fish Schools Using YOLO and Faster r-CNN on Echosounder Data.'' Proceedings of IEEE OCEANS

    2020

  19. [19]

    McQuinn, Ian H, and D Reid. 2022. ``Description of the ICES HAC Standard Data Exchange Format, Version 1.60.''

    2022

  20. [20]

    Pala, Ahmet, Anna Oleynik, Ketil Malde, and Nils Olav Handegard. 2024. ``Self-Supervised Feature Learning for Acoustic Data Analysis.'' Ecological Informatics 84: 102878

    2024

  21. [21]

    Redmon, Joseph, Santosh Divvala, Ross Girshick, and Ali Farhadi. 2016. ``You Only Look Once: Unified, Real-Time Object Detection.'' In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 779--88

    2016

  22. [22]

    Redmon, Joseph, and Ali Farhadi. 2017. ``YOLO9000: Better, Faster, Stronger.'' In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 7263--71

    2017

  23. [23]

    Ren, Shaoqing, Kaiming He, Ross Girshick, and Jian Sun. 2016. ``Faster r-CNN: Towards Real-Time Object Detection with Region Proposal Networks.'' IEEE Transactions on Pattern Analysis and Machine Intelligence 39 (6): 1137--49

    2016

  24. [24]

    Rezvanifar, Ashkan et al. 2019. ``Deep Convolutional Networks for Marine Acoustic Classification.'' IEEE Oceans Conference

    2019

  25. [25]

    Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. 2015. ``U-Net: Convolutional Networks for Biomedical Image Segmentation.'' In International Conference on Medical Image Computing and Computer-Assisted Intervention, 234--41. Springer

    2015

  26. [26]

    Schmidhuber, Jürgen. 2022. ``Scientific Integrity and the History of Deep Learning: The 2021 Turing Lecture, and the 2018 Turing Award.'' Technical Report IDSIA-77-21 (v3), IDSIA, Lugano, Switzerland, 2021--2022

    2022

  27. [27]

    Szegedy, Christian, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. 2013. ``Intriguing Properties of Neural Networks.'' arXiv Preprint arXiv:1312.6199

    Pith/arXiv arXiv 2013

  28. [28]

    Yassir, Anas, Said Jai Andaloussi, Ouail Ouchetto, Kamal Mamza, and Mansour Serghini. 2023. ``Acoustic Fish Species Identification Using Deep Learning and Machine Learning Algorithms: A Systematic Review.'' Fisheries Research 266: 106790. CSLReferences document

    2023