SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles
Pack-ice seals are key indicator species in the Southern Ocean. Their large size (2–4 m) and continent-wide distribution make them ideal candidates for monitoring programs via very-high-resolution satellite imagery. The sheer volume of imagery required, however, hampers our ability to rely on manual...
Published in: | Remote Sensing |
---|---|
Main Authors: | , , |
Format: | Text |
Language: | English |
Published: |
Multidisciplinary Digital Publishing Institute
2022
|
Subjects: | |
Online Access: | https://doi.org/10.3390/rs14225655 |
id |
ftmdpi:oai:mdpi.com:/2072-4292/14/22/5655/ |
---|---|
record_format |
openpolar |
spelling |
ftmdpi:oai:mdpi.com:/2072-4292/14/22/5655/ 2023-08-20T04:01:04+02:00 SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles Bento C. Gonçalves Michael Wethington Heather J. Lynch agris 2022-11-09 application/pdf https://doi.org/10.3390/rs14225655 EN eng Multidisciplinary Digital Publishing Institute Remote Sensing Image Processing https://dx.doi.org/10.3390/rs14225655 https://creativecommons.org/licenses/by/4.0/ Remote Sensing; Volume 14; Issue 22; Pages: 5655 pack-ice seal remote sensing Worldview-3 Antarctica computer vision deep learning instance segmentation U-Net Text 2022 ftmdpi https://doi.org/10.3390/rs14225655 2023-08-01T07:15:37Z Pack-ice seals are key indicator species in the Southern Ocean. Their large size (2–4 m) and continent-wide distribution make them ideal candidates for monitoring programs via very-high-resolution satellite imagery. The sheer volume of imagery required, however, hampers our ability to rely on manual annotation alone. Here, we present SealNet 2.0, a fully automated approach to seal detection that couples a sea ice segmentation model to find potential seal habitats with an ensemble of semantic segmentation convolutional neural network models for seal detection. Our best ensemble attains 0.806 precision and 0.640 recall on an out-of-sample test dataset, surpassing two trained human observers. Built upon the original SealNet, it outperforms its predecessor by using annotation datasets focused on sea ice only, a comprehensive hyperparameter study leveraging substantial high-performance computing resources, and post-processing through regression head outputs and segmentation head logits at predicted seal locations. Even with a simplified version of our ensemble model, using AI predictions as a guide dramatically boosted the precision and recall of two human experts, showing potential as a training device for novice seal annotators. Like human observers, the performance of our automated approach deteriorates with terrain ruggedness, highlighting the need for statistical treatment to draw global population estimates from AI output. Text Antarc* Antarctica Sea ice Southern Ocean MDPI Open Access Publishing Southern Ocean Remote Sensing 14 22 5655 |
institution |
Open Polar |
collection |
MDPI Open Access Publishing |
op_collection_id |
ftmdpi |
language |
English |
topic |
pack-ice seal remote sensing Worldview-3 Antarctica computer vision deep learning instance segmentation U-Net |
spellingShingle |
pack-ice seal remote sensing Worldview-3 Antarctica computer vision deep learning instance segmentation U-Net Bento C. Gonçalves Michael Wethington Heather J. Lynch SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
topic_facet |
pack-ice seal remote sensing Worldview-3 Antarctica computer vision deep learning instance segmentation U-Net |
description |
Pack-ice seals are key indicator species in the Southern Ocean. Their large size (2–4 m) and continent-wide distribution make them ideal candidates for monitoring programs via very-high-resolution satellite imagery. The sheer volume of imagery required, however, hampers our ability to rely on manual annotation alone. Here, we present SealNet 2.0, a fully automated approach to seal detection that couples a sea ice segmentation model to find potential seal habitats with an ensemble of semantic segmentation convolutional neural network models for seal detection. Our best ensemble attains 0.806 precision and 0.640 recall on an out-of-sample test dataset, surpassing two trained human observers. Built upon the original SealNet, it outperforms its predecessor by using annotation datasets focused on sea ice only, a comprehensive hyperparameter study leveraging substantial high-performance computing resources, and post-processing through regression head outputs and segmentation head logits at predicted seal locations. Even with a simplified version of our ensemble model, using AI predictions as a guide dramatically boosted the precision and recall of two human experts, showing potential as a training device for novice seal annotators. Like human observers, the performance of our automated approach deteriorates with terrain ruggedness, highlighting the need for statistical treatment to draw global population estimates from AI output. |
format |
Text |
author |
Bento C. Gonçalves Michael Wethington Heather J. Lynch |
author_facet |
Bento C. Gonçalves Michael Wethington Heather J. Lynch |
author_sort |
Bento C. Gonçalves |
title |
SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
title_short |
SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
title_full |
SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
title_fullStr |
SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
title_full_unstemmed |
SealNet 2.0: Human-Level Fully-Automated Pack-Ice Seal Detection in Very-High-Resolution Satellite Imagery with CNN Model Ensembles |
title_sort |
sealnet 2.0: human-level fully-automated pack-ice seal detection in very-high-resolution satellite imagery with cnn model ensembles |
publisher |
Multidisciplinary Digital Publishing Institute |
publishDate |
2022 |
url |
https://doi.org/10.3390/rs14225655 |
op_coverage |
agris |
geographic |
Southern Ocean |
geographic_facet |
Southern Ocean |
genre |
Antarc* Antarctica Sea ice Southern Ocean |
genre_facet |
Antarc* Antarctica Sea ice Southern Ocean |
op_source |
Remote Sensing; Volume 14; Issue 22; Pages: 5655 |
op_relation |
Remote Sensing Image Processing https://dx.doi.org/10.3390/rs14225655 |
op_rights |
https://creativecommons.org/licenses/by/4.0/ |
op_doi |
https://doi.org/10.3390/rs14225655 |
container_title |
Remote Sensing |
container_volume |
14 |
container_issue |
22 |
container_start_page |
5655 |
_version_ |
1774722455146856448 |