Automated aging of fish otoliths using machine learning algorithms

Project Outline:

Much effort in sustainable fisheries management is directed at estimating stock abundance. Catch and catch rates can provide some information, but inclusion of age structure information, of recreationally and commercially caught species, in stock assessment models is considered as the ‘gold standard’. To collect age data, growth rings are counted on the fish’s otolith (ear bone) in a manner similar to aging a tree by its growth rings. Aging of otoliths requires training readers on test sets of otoliths and comparing counts between reads and between readers Once trained, human readers must sit down and read hundreds if not thousands of otoliths for the species of interest. This is time consuming, costly and the repetitive nature of the task opens itself to errors through boredom and technique drift. Such a task however, is a perfect candidate for employing machine learning techniques in the development of an automated approach using images of previously collected and aged otoliths.

Hence, the aim of this project is to investigate and develop automated approaches incorporating deep learning for estimating otolith age of different species of fish. The proposed approach can be utilised in real world stock assessments, potentially providing reliable and consistent estimates and addressing limitations typically associated with human-centred methods.

Desired skills: Knowledge of Machine Learning; Experience in deep learning would be advantageous. Very strong programming skills

Project Area: Computer Vision, Image Processing, Deep Learning, Machine learning

Supervisor(s):  A/Prof. Chiou Peng Lam, A/Prof Martin Masek, Dr Rodney Duffy

Project level: Masters, PhD

Funding: Applicant to apply for ECUHDR or RTP Scholarship

Start date: Semester 2, 2020