Automated Fish Identification (AFID)

Credit: Marrable, 2020

Client / Partner
Partners: Harrier Project Management, In-Situ Marine Optics, SeaGIS, Business Research and Innovation Initiative (BRII)
Funded by: Harrier Project Management

Timeline
October 2021 – June 2023 (Completed)

Overview
Australia’s fisheries industry is valued at over $2.5 billion annually, providing a vital food source and sociocultural resource. Monitoring fish populations is essential to ensuring sustainability, but traditional methods, such as manual analysis of Baited Remote Underwater Video Systems (BRUVS) are time consuming, labour intensive and costly. This created a clear need to automate the process of fish identification and measurement to improve efficiency, reduce operational costs and enhance the scalability of monitoring programs.

The Curtin Institute for Data Science collaborated with leading marine technology and ecological monitoring partners to develop a ML based solution integrated with EventMeasure, a widely used video analysis platform. The project team, which included software engineers, marine scientists and data scientists designed deep learning models capable of detecting and identifying fish species in underwater footage, automatically measuring fish length and integrating AI outputs seamlessly into existing user workflows. The approach combined supervised learning methods, computer vision and user-driven feedback loops, refining the system through iterative testing with end users.

The outcome was a working proof of concept for the Automated Fish Identification (AFID) system. This included the development of a fish detector, an automated length measurement tool and a species identification model integrated with EventMeasure, along with multiple fish tracking capabilities, MaxN discovery, bulk labelling functionality and comprehensive data management tools. The system was tested by 30 users across seven organisations in Australia and three international partners, with feedback leading to improved algorithms. Delivered as a web based solution, the system demonstrated significant time savings compared to manual annotation, resulting in substantial reductions in labour costs.

The project led to two peer-reviewed publications and generated written expressions of interest from potential users. AFID is a product that can be licensed, making it available for adoption by organisations seeking to improve their fisheries monitoring programs.

Further information about AFID can be found at https://www.afid.io/home