FishID

The Challenge

Current methods of monitoring fish populations involve high costs for manually processing and extracting data from underwater cameras. The FishID project aims to transform environmental monitoring of aquatic ecosystems in Australia through automated detection and identification of animals in underwater imagery.

The Response

The FishID platform will overcome the cost associated with manually processing and extracting data from underwater cameras by creating a user-friendly, public-facing end-to-end pipeline for deep learning detection and automated identification of animals.

FishID will deliver a robust and intuitive system for researchers to annotate imagery, train and evaluate deep learning models to accurately detect, identify and count species of interest across coastal and marine ecosystems.

The project involves the following elements:

• An automated, integrated service for object detection and classification of underwater imagery including an annotation tool, video analysis and output, improved user interface and public API.
• Training – Videos, documentation and full training packages to support uptake and use.
• Community – Software testing with key users, and workshops with the user community.

Who Will Benefit

Researchers, federal and state government, farmers, environmental educators and tourism operators will benefit from the project’s core features:

• Integrated end-to-end automation service – A pipeline of software packages that assist annotations using Machine Learning (ML) and support model creation and testing of deep learning models for object detection and classification. Provides species identification and abundance data outputted in common database formats (e.g. for GlobalArchive).
• Provide online training – Comprehensive training package designed as user-friendly tutorials for topics including annotation, model development and model evaluation.
• Community building – Sharing knowledge among the network of Australian experts through regular workshops with research data scientists and aquatic ecosystem scientists. Enabling users to access and use models and workflows by making available code through easy-to-understand JupyterLab Notebooks. These are likely to be deployed and run in the EcoCommons command-line environment, EcoCloud.

The Partners

Our partners are:

• Griffith University
• The University of Adelaide
• University of the Sunshine Coast
• The University of Western Australia
• James Cook University
• Education Queensland (Moreton Bay Environmental Education Centre)
• Western Australia Department of Primary Industries and Regional Development

Target Outcomes

This project fills a critical infrastructure gap between the collection of underwater imagery and the resulting data made available in national repositories.

The project will enable a step-change in monitoring efficiency that will improve outcomes across multiple sectors such as:

• Marine environmental monitoring – Includes environmental assessment and State of Environment reporting.
• River health monitoring – Includes Murray-Darling Basin Authority.
• Fisheries assessments – Includes State Fisheries departments.
• Aquatic habitat restoration – NGOs investing $130 million in reef restoration over 5 years. Some already have streaming cameras suitable for automated image analysis.
• Education – Includes Moreton Bay Live streaming cameras by Queensland’s Environmental Education Centre.
• Tourism – Includes Great Barrier Reef live streaming cameras with Cairns Aquarium.

More Information

For more information on FishID, visit the project website.