One fish, two fish, red fish, blue fish: Species recognition software for sustainable fisheries

By Chris Gerbing October 17th, 2019

Technology adapted from the security industry is being taken to the oceans to monitor fishing operations, manage the seafood supply chain and ensure the sustainability of our fisheries.
Drone image of a fishing vessel in port

A long-line fishing vessel leaving port bound for the Australian fishery around Heard and McDonald Islands in the sub-Antarctic. Electronic monitoring systems are being trialled on lengthy voyages to record fishing operations. Source: Theo Verios, Austral Fisheries

Swimming in schools far below the surface is a valuable and renewable resource. Fish in great numbers support vast marine ecosystems and our growing population too. Millions of people the world over depend on fisheries for food, nutrition, income and livelihood.

A new automated visual detection tool is being developed by CSIRO’s Marine Visual Technologies group, a team originating from the ON Accelerate program, to help manage large-scale fishery sustainably, and support seafood producers to manage the supply of their product from boat-to-plate. Co-Lead scientist on the project, Dr Geoff Tuck, is excited to see the realisation of a timely solution for an industry quickly adopting cameras on vessels.

Outside of MVT, Dr Tuck heads up the team at CSIRO responsible for calculating annual catch levels for Australia’s southern and eastern scalefish and shark fishery. By monitoring fish populations limits are set each year on how many fish commercial vessels are allowed to take so that our fisheries are sustainable and profitable.

Skippers of fishing vessels are required to report the amount of each species caught to the Australian Fisheries Management Authority (AFMA), which regulates commercial fishing operations in the 8 million square kilometres of open ocean that is the Australian Fishing Zone. Further afield, international agreements also require catches to be monitored and set at sustainable levels.

From Dr Tuck and colleagues’ work with Australian fisheries, Marine Visual Technologies (MVT) was born. Bringing together expertise in fisheries management, marine conservation, ecology, software engineering and global marketing, the MVT team has been developing a software based on Artificial Intelligence (AI) technology to automatically identify what is being caught and store the information for future analysis. The application doesn’t stop at setting sustainable catch levels, but also offers seafood producers the ability to monitor and manage their product throughout the supply chain.

AI software is already used in security and surveillance for facial recognition, sports analytics and even social media. Dr Tuck believes that similar software could save industry and regulators time and money “because currently on-land technicians need to watch thousands of hours of vessel footage, and with advancements in machine learning algorithms, the process can be automated”.

Collecting good data is key to determining sustainable fishing levels. In the past human observers have been employed to count and measure the catch and un-intended bycatch coming onto a boat. This can be both costly and dangerous.

Automated eyes for better fisheries management from CSIRO on Vimeo.

“More and more around the world, human observers are being replaced with high-definition cameras,” Dr Tuck explains.

One quarter of Australian fishing vessels already use cameras in an electronic monitoring system[1] that links cameras to GPS and motion sensors, to oversee how the catch is handled, what is discarded and regrettably, what other marine life is affected. Thus, with these electronic eyes on board, there is better coverage of fishing operations at sea – but only if someone reviews the footage.

MVT can review this footage automatically. The software is nicknamed SNAPPER and was developed in collaboration with Dr Dadong Wang Team Leader at CSIRO’s Data 61 Quantitative Imaging Research Team. SNAPPER identifies and bookmarks important snippets of footage for humans to verify, thus condensing eight hours of video into eight minutes. “It means that we can use machine learning to view 100% of fishing operations at a highly reduced cost for fishers,” says Dr Wang.

One step further, the software can also recognise different species of fish. MVT’s WANDA software has the potential to count the catch by species, including any unintended by-catch, and generates a report. By tagging the fish at this stage MVT are also able to store the image and conditions of capture in a database, which can track the product through the supply chain from boat-to-plate.

Tracking fish from boat to plate from CSIRO on Vimeo.

Dr Tuck says that there is strong interest in automation products that offer cost-efficient video analytics, particularly from commercial long-line fishers. WANDA is particularly applicable to these operations, in which a central line is cast into the water with baited hooks and fish are hauled in one-by-one.

Austral Fisheries operates around Heard Island and McDonald Islands in the sub-Antarctic – the location of the valuable Patagonian toothfish fishery. Their fishing fleet makes the 4000-kilometre, 100-day voyage multiple times a year.

As they search for Patagonian toothfish, they “have two observers per trip – every single trip, every time we go there,” says Martin Exel, General Manager of Environment and Policy at Austral Fisheries. “For us, the cost of observers is extremely high so the efficacy of MVT would be excellent”.

Rough seas and two people on a boat

Long-line fishers at work in rough seas. E-monitoring systems are being trialled to better utilise human observers on board who are currently tasked to count catch and document by-catch in conditions that can be dangerous. Source: Rory Stevens, Austral Fisheries

Australian fisheries are in good health[2] thanks to stringent catch limits and good regulatory compliance1, but MVT have a global outlook. New Zealand and Chile will be installing electronic monitoring systems on thousands of fishing vessels over the next two years, and MVT products could enable the efficient monitoring of these and other large fisheries. Southeast Asian fisheries alone consist of 3.5 million fishing vessels[3], and represent a potential global market for this innovation.

Dr Tuck also believes environmentally conscious consumers will value and benefit from this technology. “With on-vessel cameras and MVT’s WANDA product, seafood can be traced throughout the supply chain, assuring consumers that the product they buy is the product they think they are buying”.

Better data means better management of our fish stocks. “We are providing products for industry, managers and conservation groups that gives them more surety about the fish that are being caught,” says Dr Tuck.

With our ever-growing demand for food, Marine Visual Technologies can help keep global fishing operations in check and our fisheries in good health, so we can all have a robust resource for the future.

Footnotes

[1] AFMA Annual Report 2015-2016. Australian Fisheries Management Authority, Canberra

[2] Status of Australian Fish Stocks Reports 2016. Fisheries Research and Development Corporation, Canberra

[3] The State of World Fisheries and Aquaculture 2016: Contributing to food security and nutrition for all. Food and Agriculture Organization of the United Nations

0 comments

Leave a Reply

Your email address will not be published. Required fields are marked *