Complex data labeling with deep learning methods: Lessons from fisheries acoustics. (March 2021)

• Record Type: Journal Article
• Authors: Sarr, Jean-Michel A.; Brochier, Timothée; Brehmer, P.; Perrot, Y.; Bah, A.; Sarré, A.; Jeyid, M.A.; Sidibeh, M.; El Ayoubi, S.
• Is Part Of: ISA transactions. Volume 109(2021)
• Page Start: 113
• Page End: 125
• Publication Date: 2021-03
• Subjects: Fisheries acoustics -- Machine learning -- Neural network -- Active acoustics -- Labeling process -- Bottom correction; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2020.09.018
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: 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. Highlights: Non-uniform expert labeling of complex data is a common issue, typically illustrated in fisheries acoustics and noisy raw datasets. A common labeling process is a first step toward comparable computational methods in fisheries acoustics. A method to automate and standardize the labeling process in fisheries acoustics with machine learning is presented. Convolution neural networks are identified as good features extractor and can benefit from non stationary fisheries acoustics datasets.