Friday, March 29, 2024

AI advancing for abalone: FRDC

Recent stories

Michelle Daw, Fisheries Research and Development Corporation (FRDC), 25 November 2022

Technology that will save time and money, provide more accurate data more often, and reduce stress on farmed abalone, is continuing to prove its worth in trials.

A multi-disciplinary team at James Cook University (JCU) is testing tools and methodology to use Artificial Intelligence (AI) for the critical task of counting and measuring abalone in grow-out tanks in aquaculture farms, under an FRDC-funded project.  

The project focusses on a “hands-off” method of counting and measuring abalone in grow-out tanks, using AI to analyse images of blacklip, greenlip and hybrid abalone, which are captured by cameras mounted on tripods. 

The research originated when James Cook University researcher, Phoebe Arbon, picked up on demand from abalone farmers for better ways to count and measure abalone. As an honours student, Phoebe developed a proposal to investigate the use of AI in abalone aquaculture and was one of the winners of the 2020 Science and Innovation awards, sponsored by FRDC.  Her initial research then led to the current FRDC project, 2019-151 – Application of a machine learning approach for effective stock management of farmed abalone. The project is funded by the Australian Abalone Growers Association and by FRDC on behalf of the Australian Government. 

The project is led by Professor Jan Strugnell, Director of the Centre for Sustainable Tropical Fisheries and Aquaculture at JCU. She explains that because abalone attach to substrates with a muscular foot that acts like a suction cap, the current on-farm practice of detaching them is time-consuming. 

“Every time you want to measure or check the abalone you need to remove them from their tank and that’s an added stressor to the abalone,” Jan says. “This handling stress can make them more susceptible to disease and can reduce growth and survival. 

Data collection on farms is limited because taking these measurements is very time consuming. 

“Currently the abalone industry spends in the order of $25,000 per annum, per farm, gathering numbers and size distributions of abalone across production.  

The AI tool will enable cost savings and produce higher quality data in greater quantities, that can be gathered more efficiently and enable better farm management decisions. “ 

Professor Ickjai (Jai) Lee, Head, Discipline of Information Technology at JCU, is working with Jan and Phoebe to teach the AI system to interpret the visual data collected via images, based on enormous data sets.  

“Over the past year, we have got the AI system to the stage where it can count five-week-old nursery stage abalone, which are about 1 millimetre in diameter, to an accuracy of over 95 per cent,” he says. 

“For abalone that are 30mm or more in diameter, we are getting 93 per cent accuracy Professor Lee says the system is also robust enough to discount non-target species which may be included in the images, such as copepods, the small crustaceans that are found in most saltwater and freshwater habitats. 

He says sampling strategies in the trial vary, depending on the data required.  

“For instance, where size is most important, one to three images per tank  or cohort may be enough to give an approximation for average size and size distribution but heavier sampling may be needed for activities such as feed trials.  

Jai says there is immense potential for further improvements and extension to the system’s capabilities. 

“Some immediately possible applications could include automating the image capture process via the usage of digital markers (Differential Global Positioning Systems) for automatic image or tank allocation,” he says. 

“Another option could be upgrading existing automated feeding or cleaning systems to enable them to capture images. These technologies would allow for much faster and labour efficient collection of information for most of the abalone growth cycle. 

“I believe long-term developments of this system will enable it to provide seamless, up to date farm knowledge at any time, allowing for more robust and cost-effective farm management.” 

And Jan says in future, AI tools could go beyond the analysis of images to create predictive capacity for outcomes such as disease outbreaks, based on a range of range of data types, such as water temperature and dissolved oxygen levels,  

“This would help farmers to potentially manipulate conditions in advance to prevent disease events.” 

This article is copyright of the Fisheries Research and Development Corporation and was published under a CC BY 3.0 AU licence.

KEEP IN TOUCH

Sign up for updates from Australian Rural & Regional News

Manage your subscription

We don’t spam! Read our privacy policy for more info.