Towards sensor agnostic artificial intelligence for underwater imagery
Towards sensor agnostic artificial intelligence for underwater imagery
Underwater images in global datasets are gathered by different cameras and under different lighting and altitude conditions. Image formation models can potentially compensate for these known differences and improve machine learning (ML) performance. In this paper we will investigate how ML trained on images from one system can classify images taken by another.
We will train two ML classification models based on two different underwater camera systems. We are going to evaluate the performance of each model with data from both platforms and will demonstrate the improvement of using image formation models to make an image look-like it has been gathered with a different system and how ML performance is affected.
Massot Campos, Miguel
a55d7b32-c097-4adf-9483-16bbf07f9120
Yamada, Takaki
81c66c35-0e2b-4342-80fa-cbee6ff9ce5f
Thornton, Blair
8293beb5-c083-47e3-b5f0-d9c3cee14be9
6 March 2023
Massot Campos, Miguel
a55d7b32-c097-4adf-9483-16bbf07f9120
Yamada, Takaki
81c66c35-0e2b-4342-80fa-cbee6ff9ce5f
Thornton, Blair
8293beb5-c083-47e3-b5f0-d9c3cee14be9
Massot Campos, Miguel, Yamada, Takaki and Thornton, Blair
(2023)
Towards sensor agnostic artificial intelligence for underwater imagery.
In International Symposium on Underwater Technology (UT23).
Record type:
Conference or Workshop Item
(Paper)
Abstract
Underwater images in global datasets are gathered by different cameras and under different lighting and altitude conditions. Image formation models can potentially compensate for these known differences and improve machine learning (ML) performance. In this paper we will investigate how ML trained on images from one system can classify images taken by another.
We will train two ML classification models based on two different underwater camera systems. We are going to evaluate the performance of each model with data from both platforms and will demonstrate the improvement of using image formation models to make an image look-like it has been gathered with a different system and how ML performance is affected.
This record has no associated files available for download.
More information
e-pub ahead of print date: 6 March 2023
Published date: 6 March 2023
Venue - Dates:
2023 IEEE Underwater Technology Symposium, Institute of Industrial Science, Tokyo, Japan, 2023-03-06 - 2023-03-09
Identifiers
Local EPrints ID: 473778
URI: http://eprints.soton.ac.uk/id/eprint/473778
PURE UUID: 3f29b201-1c31-4114-9966-72ec7dc683da
Catalogue record
Date deposited: 31 Jan 2023 17:44
Last modified: 12 Aug 2023 01:54
Export record
Contributors
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