Recognizing Ecological Behavior Patterns with Deep Learning.

PhD-candidate: Håkon Måløy

Supervisors:
Professor Keith L. Downing (NTNU)
Associate Professor Kerstin Bach (NTNU)
Senior Research Scientist Ekrem Misimi (SINTEF Ocean)

Duration: Q2 2018 – Q2 2022


Fish biology in exposed aquaculture is a four year project involving IMR, NTNU, SINTEF Ocean, Cermaq, Salmar, Marine Harvest and Kongsberg Maritim. The project aims to evaluate fundamental welfare indicators in Atlantic salmon both at the individual and school levels.

Fish biology in exposed aquaculture aims to uncover and validate operational welfare indicators for use within breeding cages. The project is expected to result in increased insight into how production affects salmon welfare as well as improved observational methods and tools.

Through the use of datadriven methods and machine learning, sensor data from within breeding cages will be analyzed to uncover co-relations between external environmental factors and the behavioral patterns in salmon.

Previous experiments using similar methods have resulted in promising results [1] and access to more and better sensor data makes it possible to test and evaluate candidate approaches in lab. The most promising approaches will then be tested in large scale, operational environments.

Results so far:

  • The PhD project started in april 2018. The main focus has been on completing obligatory courses, literature searches and required duties. There are therefore no results to show as of 29.01.2019.

Status and future work:

  • Currently sensor data is being collected for initial analysis and preliminary experiments.
  • In 2019 the project will focus on continued completion of obligatory courses as well as experiments in the lab.
  • The most promising approaches will be tested at full scale production environments in cages towards the end of the project.

[1] Måløy, H., Aamodt, Agnar, & Misimi, Ekrem. (2017). A Dual-Stream Deep Learning Architecture for Action Recognition in Salmon from Underwater Video. https://brage.bibsys.no/xmlui/handle/11250/2502728 .