Stereoscopic cameras could soon be researchers’ eyes at sea, with trials to collect Giant Crab (Pseudocarcinus gigas) data improving stock assessments for other species too.
Simulating human vision to capture 3D images through stereoscopic cameras could soon be playing an important role in gathering information on Giant Crabs (Pseudocarcinus gigas).
By capturing Giant Crab images and using artificial intelligence (AI) to gauge length, growth, and sex data, researchers will gain access to key characteristics needed for quality stock assessments – data that has been hard to gather.
These cameras will be trialled on board vessels in the Giant Crab Fishery by the end of the year.
The trial of on-board cameras in the fishery initially began with a standard camera set-up, says Toby Jeavons, Estuarine Fisheries Manager with the Victorian Fisheries Authority, who is leading this FRDC-funded project (2019-114).
The system had a single camera attached to a Raspberry PI minicomputer, a portable hard drive backup, and a miniature screen.
“The idea is that you place the crab on a special sorting table, under the camera, and take a picture using a foot pedal,” says Toby.
“The image is captured and then stored. Back on land, the images are sent to us, and AI works its magic. It analyses all the images and works out length, sex and ideally crab recapture data too.”
Although this initial trial worked, the team, which includes the Institute for Marine and Antarctic Studies (IMAS) at the University of Tasmania, found there could be improvements.
Via a project extension, the team is now trialling a stereoscopic camera system, with two cameras. This simulates human binocular vision and generates 3D images, providing greater accuracy when measuring length.
“It also increases the speed at which we’re able to take images and improves image quality. That then improves the accuracy of what we can feed into models,” explains Toby.
He says the team at the IMAS team led by Dr Scott Hadley, is changing the game for on-board data collection systems. The next round of on-board industry testing will occur towards the end of 2023, providing the team with further feedback.
Plugging the data gap
Quality stock information on Giant Crabs has been hard to collect. These crabs are a long-living, slow-reproducing species found in southern Australian waters from central NSW to southern Western Australia, including Tasmania.
Stock assessments have relied largely on fisher-dependent catch rate data. However, the remote nature of fishing operations and the small number of operators makes it difficult to obtain independent observer coverage. This creates an inherent level of volatility in the data.
Attempts to improve fishery data collection have been challenging, particularly for length-frequency data collected by fisher-based volunteer programs.
Unfortunately, this lack of data has come at a time when it is needed most. Despite harvest strategies and management plans, trends show Giant Crab stocks are declining.
Toby says the first step in improving the robustness of stock assessments is quality data and that needs innovative collection methods.
Previous length-based models were developed to integrate assessments across state jurisdictions. “However, data limitations meant it was never used in Victoria and insufficient length-frequency data has seen it discontinued in South Australia and Tasmania.
“We’ve been missing all that length-frequency data which has meant we have been unable to run biomass modelling estimates.”
Stock assessment challenges
“Finding a solution to address this knowledge gap is really important to improve the robustness of stock assessments.,” explains Toby.
Through this project, we can start to collect this data again and work towards integrating assessments across jurisdictions at the whole of stock level.”
FRDC Project Manager Dr Toby Piddocke highlights the need for the regular and repeated collection of high-quality data to inform stock assessments as an ongoing challenge in fisheries science.
“This is a particularly pressing issue in small-scale fisheries, where sustained resourcing for data collection is often difficult to obtain,” he says.
“Digital image recognition and AI, like that being developed in this project, offer considerable potential to develop and sustain these data streams. It’s why FRDC has invested in this work.”
As the IMAS team leader on the project, Dr Scott Hadley suggests the stereoscopic imaging system could be ideal for a range of small, high-value fisheries.
“You can control the environment and capture images with minimal gear. It might not be best for larger fisheries where you can have 50 different types of bycatch on a deck that look different every time, but for this type [of small-scale fisheries], it works well,” Scott says.
Shell pattern identifiers
Future plans include identifying recaptured crabs using AI to find each crab’s unique shell pattern.
This “fingerprint” equivalent would allow researchers to better understand animal growth, feeding important information into stock assessment models.
The technical challenge however is Giant Crabs’ inter-moult period – the time between the shedding of its shell as the crab grows – which is one of the longest of any crab species. It can be up to 15 years between moults for mature females, although it is believed the unique shell pattern remains.
Given the long intermoult, IMAS is testing its methodology using Southern Rock Lobsters (Jasus edwardsii) to verify that shell patterns remain post-moult. Challenges for the imaging of rock lobsters include biofouling of the shells and possible distortion of the shell patterns, due to growth.
Success in that trial will inform how the imaging is applied to recaptured Giant Crabs.
Toby Jeavons says that the team will also continue trying to identify recaptured Giant Crabs. “The hope would be that we can identify recaptured individuals and measure growth between those intervals, to determine growth rates,” he says.
The project’s innovative collection and AI methods has already been extended to the Crystal Crab Fishery in Western Australia, with some tweaks. It is a higher volume fishery, which means more processing data of more animals so still images and video is being collected for analysis.
Innovative data collection methods are evolving and will improve how important commercial fisheries stocks are monitored. “At the end of the day, approaches to collecting data that have minimum burden and cost to fishers will enhance the long-term sustainability of these resources,” says Toby.
2019-114: Giant Crab Enhanced Data Collection - Innovative approaches to enhance data collection in the Victorian, South Australian and Tasmanian Giant Crab fisheries.
Off King Island and onboard the Western Light, IMAS Technical Officer Ben Quigley uses a new image-capturing system to photograph Giant Crabs. Operated by a foot pedal, the system captures key crab characteristics. Via AI, this information is then used to gauge length, growth, and sex data, which is vitally needed for robust stock assessments. The system is designed to be easy to use and will help plug a data gap.
Photos: Supplied by Scott Hadley, IMAS