Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BenthicNet: A global compilation of seafloor images for deep learning applications

Scott C. Lowe, Benjamin Misiuk, Isaac Xu, Shakhboz Abdulazizov, Amit R. Baroi, Alex C. Bastos, Merlin Best, Vicki Ferrini, Ariell Friedman, Deborah Hart, Ove Hoegh-Guldberg, Daniel Ierodiaconou, Julia Mackin-McLaughlin, Kathryn Markey, Pedro S. Menandro, Jacquomo Monk, Shreya Nemani, John O'Brien, Elizabeth Oh, Luba Y. Reshitnyk, Katleen Robert, Chris M. Roelfsema, Jessica A. Sameoto, Alexandre C. G. Schimel, Jordan A. Thomson, Brittany R. Wilson, Melisa C. Wong, Craig J. Brown, Thomas Trappenberg

TL;DR

BenthicNet tackles the bottleneck in automated benthic habitat analysis by compiling a global, multi-source seafloor image collection, enabling robust deep learning with both large unlabelled and a substantial labelled subset translated to CATAMI and WoRMS taxonomy.The approach combines extensive data curation, standardized metadata, and a self-supervised learning (SSL) pretraining regime on unlabelled images to improve transfer to downstream habitat-labeling tasks, demonstrated on substrate and benthoscape classification benchmarks.Key contributions include the 11.4M unlabelled images, 1.88M labelled images, 3.1M CATAMI annotations, a 1M-subset for SSL pretraining, and openly available data and model weights to enable broad reuse and benchmarking.The work highlights practical implications for global benthic habitat mapping by providing a scalable, reproducible data backbone and demonstrating SSL-based transfer advantages, especially for site-specific or low-label regimes.

Abstract

Advances in underwater imaging enable collection of extensive seafloor image datasets necessary for monitoring important benthic ecosystems. The ability to collect seafloor imagery has outpaced our capacity to analyze it, hindering mobilization of this crucial environmental information. Machine learning approaches provide opportunities to increase the efficiency with which seafloor imagery is analyzed, yet large and consistent datasets to support development of such approaches are scarce. Here we present BenthicNet: a global compilation of seafloor imagery designed to support the training and evaluation of large-scale image recognition models. An initial set of over 11.4 million images was collected and curated to represent a diversity of seafloor environments using a representative subset of 1.3 million images. These are accompanied by 3.1 million annotations translated to the CATAMI scheme, which span 190,000 of the images. A large deep learning model was trained on this compilation and preliminary results suggest it has utility for automating large and small-scale image analysis tasks. The compilation and model are made openly available for reuse at https://doi.org/10.20383/103.0614.

BenthicNet: A global compilation of seafloor images for deep learning applications

TL;DR

BenthicNet tackles the bottleneck in automated benthic habitat analysis by compiling a global, multi-source seafloor image collection, enabling robust deep learning with both large unlabelled and a substantial labelled subset translated to CATAMI and WoRMS taxonomy.The approach combines extensive data curation, standardized metadata, and a self-supervised learning (SSL) pretraining regime on unlabelled images to improve transfer to downstream habitat-labeling tasks, demonstrated on substrate and benthoscape classification benchmarks.Key contributions include the 11.4M unlabelled images, 1.88M labelled images, 3.1M CATAMI annotations, a 1M-subset for SSL pretraining, and openly available data and model weights to enable broad reuse and benchmarking.The work highlights practical implications for global benthic habitat mapping by providing a scalable, reproducible data backbone and demonstrating SSL-based transfer advantages, especially for site-specific or low-label regimes.

Abstract

Advances in underwater imaging enable collection of extensive seafloor image datasets necessary for monitoring important benthic ecosystems. The ability to collect seafloor imagery has outpaced our capacity to analyze it, hindering mobilization of this crucial environmental information. Machine learning approaches provide opportunities to increase the efficiency with which seafloor imagery is analyzed, yet large and consistent datasets to support development of such approaches are scarce. Here we present BenthicNet: a global compilation of seafloor imagery designed to support the training and evaluation of large-scale image recognition models. An initial set of over 11.4 million images was collected and curated to represent a diversity of seafloor environments using a representative subset of 1.3 million images. These are accompanied by 3.1 million annotations translated to the CATAMI scheme, which span 190,000 of the images. A large deep learning model was trained on this compilation and preliminary results suggest it has utility for automating large and small-scale image analysis tasks. The compilation and model are made openly available for reuse at https://doi.org/10.20383/103.0614.
Paper Structure (33 sections, 8 figures, 6 tables)

This paper contains 33 sections, 8 figures, 6 tables.

Table of Contents

  1. Background & Summary
  2. Methods
  3. Data compilation and quality control
  4. Individual contributions
  5. DFO
  6. NRCan
  7. NGU
  8. MGDS
  9. NOAA
  10. USGS
  11. USAP-DC
  12. AADC
  13. SQUIDLE+
  14. FathomNet
  15. PANGAEA
  16. ...and 18 more sections

Figures (8)

  • Figure 1: Distribution of images after spatial subsampling projected to Equal Earth. (Top) images according to data source and (bottom) aggregated by their density and scaled logarithmically.
  • Figure 2: Distribution of BenthicNet-1M images according to (a) (log scale) depth data retrieved from the GEBCO2022 grid gebco2022 and (b) year of acquisition.
  • Figure 3: Examples of BenthicNet images from each sampled Ecological Marine Unit (EMU), indicated in the top-left of each image. See sayer2017 for a full description of the EMU classes.
  • Figure 4: Distribution of BenthicNet-1M images according to bottom layer Ecological Marine Units (EMUs). (a) Proportion and area of global oceans classified into each EMU. (b) Proportion of BenthicNet image samples from each EMU. See sayer2017 for a full description of the EMU classes.
  • Figure 5: Distribution of BenthicNet-1M images according to global ocean basinssayer2017. (a) Proportion and area of ocean basins. (b) Proportion of BenthicNet image samples from each ocean basin.
  • ...and 3 more figures