Home // COCE 2025, The Second International Conference on Technologies for Marine and Coastal Ecosystems // View article

Operationalizing and Testing Machine Learning Models for Acoustic Target Classification

Authors:
Nils Olav Handegard
Silje Smith-Johnsen
Arne Johannes Holmin
Cristian Muños Mas
Ingrid Utseth
Daniel Dondorp

Keywords: acoustic trawl surveys; machine learning; micro services; containerisation; data linage; MLOps; CI/CD.

Abstract:
Acoustic trawl surveys use echosounders to collect acoustic backscatter, which is categorized and combined with trawl samples to generate abundance indices for fisheries assessment models. Machine learning models are being developed to automate the acoustic target classification step, and it is necessary to evaluate their performance in comparison to manual processes and earlier model versions. The data processing pipeline consists of several stages, utilizing various software, versions, and libraries. Docker containers provide flexibility, especially for advanced pipeline steps. Some steps use Python virtual environments. Clearly defining data models between processing steps is necessary and adopting community standards where applicable is recommended. We have set up a system to combine and run the pipeline steps, and we have used it to compare different ML models. We are currently working to further streamline the process.

Pages: 23 to 26

Copyright: Copyright (c) The Government of Norway, 2025. Used by permission to IARIA.

Publication date: October 26, 2025

Published in: conference

ISBN: 978-1-68558-329-3

Location: Barcelona, Spain

Dates: from October 26, 2025 to October 30, 2025