Cameras show real potential for catch monitoring

This story was first published in Fishing Industry News.

Monitoring the catches of the deep-sea trawl fishery with cameras is not only efficient and affordable, but it has the potential to substantially improve the management of bycatch in the fishery, a new study has found.


The cameras that were mounted in the fish factory of a freezer trawler to test the application of cameras for monitoring bycatch in the deep-sea trawl fishery

You have to admire University of Cape Town Master’s student Michelle Lee. While her fellow students were out in the field emulating Sylvia Earle, a world-famous marine biologist, Lee was sitting in a darkened room watching thousands of hake and other deep-sea species roll over the conveyor of an on-board fish factory. Lee watched videos for weeks on end, timing herself so that she could evaluate the efficiency and cost of her work, and carefully counting and identifying the species other than hake that caught the camera’s electronic eye.

Lee’s study “Electronic monitoring in the South African demersal trawl industry and its use in monitoring communities commonly caught as bycatch” will be complete in June this year, but her results have already provided ample reward for her slog. Lee has shown that a good quality camera capable of recording and processing hundreds of hours of footage, can provide more accurate estimates of the composition and volume of the catch of demersal trawlers at a lower cost than the deployment of scientific observers on fishing trips.

Her study was undertaken with Colin Attwood, associate professor of biological sciences at UCT. It was funded and supported by the South African Deep-Sea Trawling Industry Association (SADSTIA) and WWF South Africa, which awarded Lee its prestigious Master’s Fellowship Grant.

Lee describes how her study began with a very practical dilemma – finding a camera that is capable of recording high quality images for an extended period of time (fishing trips can last up to 60 days), in an environment that is unstable and wet.

“The first system we used was one that was recommended to us, but the quality of the images was poor,” relates Lee. “There were issues with exposure because the fluorescent lights in the factory impact the colour of the image, and distort the images a little bit. The motion of the conveyor belt was also a major problem.”

Bryan Fitchat and Brandon Davids of Keep Electronics, experts in security cameras and access control, helped to solve the problem. As it turned out, the solution wasn’t a high-resolution camera, but a camera with good processing speed. Another solution was very simple to implement – making the fishers on board the vessel aware of the camera and asking them to be careful of it when cleaning the factory. Having an observer on board to clean the lens on a daily basis was also extremely helpful.

Once high-quality images started rolling onto Lee’s hard drives, she began the laborious process of analysing the footage.

“For every fishing trip I would select three randomly chosen 20-minute intervals – basically an hour a day,” she explains. “I would watch that hour and manually count all the fish that came from the unsorted line and the discards line. I’d identify each individual as closely as I could and I also estimated its length, relative to the width of the conveyor belt.”

The time Lee spent on each video was roughly equivalent to the 5% to 10% coverage of the fishing fleet that the current scientific observer programme achieves.

Her results show that the percentage of hake caught by a freezer trawler in the deep-sea trawl fishery is approximately 97%.

“On one video I counted 6 000 hake in 20 minutes,” says Lee, “what I am finding is that the percentage of hake caught is 97%. It’s very high. We’re catching hake very well. We’re very good at it.”

Bycatch consists primarily of monk, kingklip and jacopever – all of which are retained, processed and marketed to very good effect by the industry – but the footage does reveal a catch of several other species that are less well used and sometimes discarded.

The advantage of the camera system is that Lee was able to precisely identify and quantify these fish, whereas on-board observers typically count them as and when they occur in their samples, and classify them very broadly into families such as, for example, “dogfish” or “catshark”. Having more precise information makes it possible for scientists and fisheries managers to understand the numbers of each species that are landed by the fishery and, if necessary, put measures in place to control them. Such measures might include, for example, “move-on” rules that compel a skipper to fish elsewhere if the percentage of a certain species in a haul exceeds agreed limits, or the closure of specific fishing zones where the ratio of that species to hake is known to be high. Restrictions such as these are already in place for monk and kingklip.

The implications of Lee’s study are far reaching, as her supervisor Attwood explains:

“For identifying and quantifying bycatch, this is better than having an observer on board,” he says. “The other advantage is that monitoring is around the clock and you can go back (to the footage) and verify a piece of information because it’s captured on video.”

“There are some types of sampling that actually involve collecting fish and with cameras that won’t be done, so one has to look at some loss of functionality. But if it weren’t for that, the cameras certainly appear to be better than observers. They can monitor three places in the fish factory simultaneously, which an observer obviously can’t do.”

Attwood cautions that the results of Lee’s study don’t necessarily mean the observer programme will close and be replaced by onboard cameras. In fact, his advice to SADSTIA has been to wait until artificial intelligence (AI) can be used to analyse the videos. A company with global experience in the use of face recognition software has already indicated that identifying and counting fish is eminently possible with AI.

“AI is more than possible,” says Attwood, “it will dramatically reduce the number of people required to watch the videos – it will all be done by machine. If that’s the case, there’s no reason why we can’t watch every part of every video. If that happens then we’ve vastly improved, to the maximum extent.”

In the short- to medium-term it’s unlikely that the industry will see human observers phased out and cameras phased in. For one thing, there’s the issue of cost. Not so very long ago, the Offshore Resources Observer Programme was fully funded by government, but today SADSTIA carries all the costs of the deployment of observers on its vessels. Will the industry be expected to fund the roll out of cameras on its vessels too? And who would manage the analysis and storage of the data that is collected?

“It’s not as if overnight there’s going to be cameras everywhere,” notes Attwood, “it will be one ship at a time and observers might find themselves working on the smaller boats where it’s more difficult to install cameras.”

The fourth re-assessment of the South African trawl fishery by the Marine Stewardship Council is currently underway and ecosystem considerations such as the management of bycatch are more important than ever. For this reason alone it is encouraging to see SADSTIA working with the scientific community and using disruptive technology to develop and improve the systems in place to monitor and manage the fishery.