Complex data labeling with deep learning methods: Lessons from fisheries acoustics

Type Article
Date 2021-03
Language English
Author(s) Sarr J.M.A.1, 2, Brochier T.1, 2, Brehmer Patrice3, 4, Perrot Yannick3, Bah A.1, 2, Sarré A.4, Jeyid M.A.5, Sidibeh M.6, El Ayoubi S.7
Affiliation(s) 1 : Université Cheikh Anta Diop de Dakar UCAD, Ecole Supérieure Polytechnique, BP 15915, Dakar Fann, Senegal
2 : IRD, Sorbonne Université, UMMISCO, F-93143, Bondy, France
3 : IRD, Univ Brest, CNRS, Ifremer, LEMAR, Plouzané, France
4 : ISRA/CRODT, Pole de recherche de Hann, BP2241, Dakar, Senegal
5 : IMROP, BP22, Nouadhibou, Mauritania
6 : Fisheries Department (FD), Marina Bay, Banjul, The Gambia
7 : INRH, Anza 80000, Agadir, Morocco
Source Isa Transactions (0019-0578) (Elsevier BV), 2021-03 , Vol. 109 , P. 113-125
DOI 10.1016/j.isatra.2020.09.018
WOS© Times Cited 9
Keyword(s) Fisheries acoustics, Machine learning, Neural network, Active acoustics, Labeling process, Bottom correction

Quantitative and qualitative analysis of acoustic backscattered signals from the seabed bottom to the sea surface is used worldwide for fish stocks assessment and marine ecosystem monitoring. Huge amounts of raw data are collected yet require tedious expert labeling. This paper focuses on a case study where the ground truth labels are non-obvious: echograms labeling, which is time-consuming and critical for the quality of fisheries and ecological analysis. We investigate how these tasks can benefit from supervised learning algorithms and demonstrate that convolutional neural networks trained with non-stationary datasets can be used to stress parts of a new dataset needing human expert correction. Further development of this approach paves the way toward a standardization of the labeling process in fisheries acoustics and is a good case study for non-obvious data labeling processes.

Full Text
File Pages Size Access
Author's final draft 48 5 MB Open access
13 2 MB Access on demand
Top of the page