Automated Detection of Antarctic Benthic Organisms in High-Resolution In Situ Imagery to Aid Biodiversity Monitoring

Cameron Trotter, Huw Griffiths, Tasnuva Ming Khan, Rowan Whittle · 2025

Read Automated Detection of Antarctic Benthic Organisms in High-Resolution In Situ Imagery to Aid Biodiversity Monitoring by Cameron Trotter, Huw Griffiths, Tasnuva Ming Khan, Rowan Whittle (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-2507-21665.txt.

Introduction

Cameron Trotter, Huw Griffiths, Tasnuva Ming Khan, and Rowan Whittle teach computers to spot Antarctic seafloor animals in photos, in a 2025 paper. Counting these creatures tracks how the ecosystem responds to climate pressure. Hand labeling the images is slow expert work.

The team releases the first public dataset of its kind from the Weddell Sea. Their detector handles huge images by cutting them into patches without shrinking. Boosted training and clever merging of patch results lifted accuracy. The system found medium and large animals across 25 body types well. Small and rare species stayed hard to catch. Still, the tool lays groundwork for machine aided ocean monitoring.

Catalog record

CREATORCameron Trotter, Huw Griffiths, Tasnuva Ming Khan, Rowan Whittle
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-21665

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Automated Detection of Antarctic Benthic Organisms in High-Resolution In Situ Imagery to Aid Biodiversity Monitoring Thanks: Corresponding author Cameron Trotter, Huw Griffiths, Tasnuva Ming Khan, Rowan Whittle Abstract Monitoring benthic biodiversity in Antarctica is vital for understanding ecological change in response to climate-driven pressures. This work is typically performed using high-resolution imagery captured in situ, though manual annotation of such data remains laborious and specialised, impeding large-scale analysis. We present a tailored object detection framework for identifying and classifying Antarctic benthic organisms in high-resolution towed camera imagery, alongside the first public computer vision dataset for benthic biodiversity monitoring in the Weddell Sea. Our approach addresses key challenges associated with marine ecological imagery, including limited annotated data, variable object sizes, and complex seafloor structure. The proposed framework combines resolution-preserving patching, spatial data augmentation, fine-tuning, and postprocessing via Slicing Aided Hyper Inference. We benchmark multiple object detection architectures and demonstrate strong performance in detecting medium and large organisms across 25 fine-grained morphotypes, significantly more than other works in this area. Detection of small and rare taxa remains a challenge, reflecting limitations in current detection architectures. Our framework provides a scalable foundation for future machine-assisted in situ benthic biodiversity monitoring research. 1 Introduction Benthic communities, comprised of organisms that live in, on or around the seafloor, are highly biodiverse, play key roles within global nutrient cycling, and are a valuable food source [ 25 ] . Global anthropogenic change, e.g., ocean warming and acidification [ 44 , 47 ] , coupled with direct local and regional pressures such as harvesting and pollution, are negatively impacting the structure and function of benthic communities [ 11 ] . The Antarctic benthos is uniquely adapted to its isolated and frozen environment [ 2 ] . These cold-adapted species face additional pressures through changes to the cryosphere that dominates their ocean, e.g., glacial melt and ice shelf collapse. These changes are most notable in the shallow benthic communities of the West Antarctic Peninsula, where changes to biodiversity, trophic structure, biomass, and distribution have been observed [ 19 ] . Historically, the exploration and monitoring of benthic environments has relied on invasive, non-quantitative methods such as dredging, or more quantitative yet slow-to-deploy instruments like corers and grabs [ 45 ] . In recent years, the adoption of imaging technologies, delivered via SCUBA, submersibles, towed or drop camera systems, remotely operated vehicles, and autonomous platforms, has significantly increased both the rate and scale of data acquisition. Photographic and video data enable rapid, in situ, and quantitative surveys of extensive seafloor areas. Imaging techniques represent a non-destructive and repeatable survey method to monitor ecosystem change. To date, the usefulness of collected data has been restricted by the need for expert assessment of every image, which is time consuming [ 52 , 3 ] and prone to fatigue and annotation bias [ 12 , 15 , 41 ] . This bottleneck is particularly evident for Antarctica, with highly diverse and endemic benthic species [ 2 ] and relatively few taxonomic experts capable of providing confident image-based identifications. Additionally, the high logistical and financial costs associated with deep-sea data collection often results in comparatively small amounts of collected data. Antarctica’s geographic isolation and extreme environmental conditions make fieldwork highly resource-intensive, limiting collection to infrequent, short-duration missions typically led by national research programmes. Recent advances in deep learning and computer vision have enabled the development of machine-assisted in situ biodiversity monitoring tools, designed to automate parts of the data curation process and mitigate the annotation bottleneck faced by marine ecologists [ 49 ] . By leveraging manually curated data from previous surveys, researchers can now train models to detect benthic organisms commonly encountered in their study regions, supporting applications such as first-pass annotation workflows [ 26 , 40 , 39 ] . Given the high data collection and annotation costs associated with the Antarctic benthos however, there is a notable lack of publicly available datasets suitable for training automated biodiversity monitoring tools for these ecosystems. This limitation is further compounded by the region’s high levels of endemism, which reduces the relevance and transferability of models trained on data from other regions. In scenarios requiring the detection of small or densely aggregated organisms, high-resolution imagery is often utilised to enable finer-scale ecological observations. However, such data introduces additional complexity to the model development pipeline. High-resolution images place substantial computational demands on both training and inference processes, and the accurate detection of small or closely packed objects remains a persistent challenge for deep learning-based object detection systems [ 30 ] . In this study, we present an object detection framework designed to identify benthic organisms in high-resolution seafloor imagery from the Weddell Sea, Antarctica. Our approach accommodates large-scale inputs without downscaling through a patch-based processing methodology. The model is trained on a manually annotated dataset of only 100 images, which we release publicly as the first computer vision–ready benthic dataset from the Weddell Sea. The resulting model is capable of detecting a wide variety of benthic organisms, more than previous works in this area and to a higher level of granularity. 2 Related Work Early studies into machine-assisted in situ benthic biodiversity monitoring used local features and hand-engineered pipelines to identify specific organisms of interest [ 13 , 48 ] . Such methods require extensive adaptation for new organisms or environments, limiting their use in broad biodiversity monitoring surveys enabled by advances in underwater imaging and affordable data storage. Data-driven deep learning models capable of generalised, automated feature extraction, e.g., Convolutional Neural Networks (CNNs), have accelerated the pace of machine-assisted in situ benthic biodiversity monitoring research [ 49 ] . Such models are often task-specific: image classifiers for taxonomic ID [ 41 , 56 ] , object detectors for abundance estimation [ 32 , 55 ] , semantic segmenters for habitat mapping and behaviour analysis [ 35 , 38 , 20 ] , and instance segmenters for biomass estimation [ 31 ] . Few studies focus specifically on the Antarctic benthos, likely due to its remoteness and high fieldwork costs. [ 32 ] make use of a YOLOv5 [ 23 ] model pre-trained on the COCO dataset [ 28 ] to provide abundance estimates for two coarse-grained morphotypes, organism groupings classified based on shared morphological characteristics, in imagery from a stationary camera deployed in the Ross Sea. As imagery is downscaled, detection of small organisms may not be feasible using the presented methodology. [ 31 ] apply a patching strategy to towed-camera imagery from the Weddell Sea to evaluate the effectiveness of synthetic data augmentation in training a CenterMask [ 27 ] model for instance segmentation of three coarse-grained morphotypes. However, their approach does not address potential detection failures at patch boundaries or support full-image ecological analysis. To address these limitations, we extend patching with overlap and postprocessing techniques that improve detection accuracy at patch edges. Additionally, we reproject detections back onto the original large-scale imagery, preserving spatial context for ecological analysis. Detailed methodology is provided in Section 4 . Further, while the aforementioned works demonstrate the feasibility of identifying Antarctic organisms using deep learning, they are limited to a small number of coarse-grained morphotypes. These studies also do not consider taxa with low abundance,

Read the complete text