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

TL;DR

This paper explores how deep learning, specifically convolutional neural networks, can assist in labeling complex fisheries acoustics data, reducing expert effort and improving data standardization.

Contribution

It demonstrates that CNNs trained on non-stationary datasets can identify parts of data needing expert correction, aiding in standardizing fisheries acoustics data labeling.

Findings

CNNs can identify data segments requiring expert review.

Deep learning reduces manual labeling effort.

Approach facilitates standardization in fisheries acoustics labeling.

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.