The effect of physics-based corrections and data augmentation on transfer learning for segmentation of benthic imagery
The effect of physics-based corrections and data augmentation on transfer learning for segmentation of benthic imagery
Ocean observation has been greatly improved by theuse of Autonomous Underwater Vehicles and Remotely OperatedVehicles, and the high quality and high quantity of imagery theyproduce. This quantity of images collected on research cruiseshas, however, become intractable using traditional manual imageanalysis methods. There is a growing need for automation of thisprocess, using methods such as deep learning to analyse and sum-marise the information present, yet research into how to improvetheir performance for underwater images is currently limited.This paper presents a study into the effect of using physics-basedcorrections and data augmentation to aid the performance ofDeepLabV3, a state of the art image segmentation system. Usingmetadata about the physical environment the images were takenin, particularly the altitude of the vehicle for each image, andthe known wavelength dependent light attenuation over distancethrough water, reduces the generalisation error of the DeepLabV3system.
Walker, Jennifer
267d3703-c27a-469e-9eb5-1480e94ea13a
Prugel-Bennett, Adam
b107a151-1751-4d8b-b8db-2c395ac4e14e
Thornton, Blair
8293beb5-c083-47e3-b5f0-d9c3cee14be9
Yamada, Takaki
81c66c35-0e2b-4342-80fa-cbee6ff9ce5f
Walker, Jennifer
267d3703-c27a-469e-9eb5-1480e94ea13a
Prugel-Bennett, Adam
b107a151-1751-4d8b-b8db-2c395ac4e14e
Thornton, Blair
8293beb5-c083-47e3-b5f0-d9c3cee14be9
Yamada, Takaki
81c66c35-0e2b-4342-80fa-cbee6ff9ce5f
Walker, Jennifer, Prugel-Bennett, Adam, Thornton, Blair and Yamada, Takaki
(2019)
The effect of physics-based corrections and data augmentation on transfer learning for segmentation of benthic imagery.
In 2019 IEEE Underwater Technology (UT).
IEEE.
8 pp
.
(doi:10.1109/UT.2019.8734463).
Record type:
Conference or Workshop Item
(Paper)
Abstract
Ocean observation has been greatly improved by theuse of Autonomous Underwater Vehicles and Remotely OperatedVehicles, and the high quality and high quantity of imagery theyproduce. This quantity of images collected on research cruiseshas, however, become intractable using traditional manual imageanalysis methods. There is a growing need for automation of thisprocess, using methods such as deep learning to analyse and sum-marise the information present, yet research into how to improvetheir performance for underwater images is currently limited.This paper presents a study into the effect of using physics-basedcorrections and data augmentation to aid the performance ofDeepLabV3, a state of the art image segmentation system. Usingmetadata about the physical environment the images were takenin, particularly the altitude of the vehicle for each image, andthe known wavelength dependent light attenuation over distancethrough water, reduces the generalisation error of the DeepLabV3system.
This record has no associated files available for download.
More information
Submitted date: March 2019
e-pub ahead of print date: April 2019
Venue - Dates:
Underwater Technology 2019, , Kaohsiung, Taiwan, 2019-04-16 - 2019-04-19
Identifiers
Local EPrints ID: 428954
URI: http://eprints.soton.ac.uk/id/eprint/428954
ISSN: 2573-3796
PURE UUID: 3774bd3e-ecac-4fc0-9a25-b8f90bc69ac3
Catalogue record
Date deposited: 15 Mar 2019 17:30
Last modified: 16 Mar 2024 04:31
Export record
Altmetrics
Contributors
Author:
Jennifer Walker
Author:
Adam Prugel-Bennett
Author:
Takaki Yamada
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics