by Simon Edwards
Traps and surveillance cameras using artificial intelligence (AI) will soon be joining the arsenal of tools available to achieve New Zealand’s ‘predator-free by 2050’ ambitions.
In a webinar last month hosted by Predator Free NZ, Dr Helen Blackie – a Biosecurity Consultant at Boffa Miskell – said field trials of AI-driven traps get underway soon. The traps, made by New Zealand technology research and development company Critter Solutions, should be commercially available by the end of next year.
While the current shortages of microchips and electronic components, compounded by “huge delays” in delivery times, make it difficult to predict final costs, “they’re not going to be more expensive than most of the other re-setting traps currently on the market, despite the fact they’ll have those AI capabilities,” Dr Blackie told the scores of webinar participants.
“What I mean by artificial intelligence – and this is just the most general definition possible – is intelligence demonstrated by machines rather than us humans. So it’s a form of machine learning so that something else can make a decision rather than us sitting there in person making the decision ourselves.”
Critter Solutions is using AI to design better control traps and detect/recognise pests.
After a lot of development work and trials, the company now has tools that enable machine learning to instantly recognise species.
“We’ve spent four or five years collecting data on a variety of species on mainland and offshore islands. For a trap, identification of the species needs to be instant, not something that happens five minutes later. It takes a long time to get these things working instantaneously.”
With funding from Predator Free 2050 last year, work has progressed on integrating the AI into a resetting trap.
The trouble with many existing traps is that their design is as much about keeping unwanted species out as trapping and killing the target pest. So they feature wire baffles, narrow tunnels or other devices to keep the likes of curious Keas, cats., etc, out.
“What that means is that we get a much lower capture success rate because we’re also making it too difficult for the target species to come in.”
Researchers have plenty of videos of pest species, stoats in particular, approaching or entering a trap and then scarpering “because they decide it’s all a bit too hard, or they get scratched by the baffles. I always say to people, make sure you file your baffles back or put a bit of tubing around it so it’s nice and smooth,” Dr Blackie said.
With the AI instant recognition, “all those constraints on the trap architecture are gone.
“We don’t need the pests to do anything. They don’t need to pull on a bite bar, they don’t need to stand on a metal plate.
“With AI we can make sure the animal is correctly placed in the trap to trigger it. Some of you will have seen in DOC traps where rats might be caught by the tail, or an animal has been caught by a leg. The AI traps are more humane and they’re safe to use in areas where you’ve got young children who can’t be trusted to not interfere with them.”
The AI traps can be ‘set and forget’ (the goal is 12 months with no servicing) and can be integrated with Critter Solution’s EzyLure dispenser. The trap mechanisms can remotely be switched on after days or weeks of pre-feeding of non-poisonous baits, a practice that Dr Blackie says has been proven to increase pest interaction – and ultimately kill rates.
There are two trap designs – one designed for mustelid/rat sized pests (but will also detect and kill mice), the other for possum-sized pests.
“The traps are capable of over 100 kills (before servicing is needed) so basically it’s resettable as many times as you’re ever going to need,” Dr Blackie said.
The other AI focus for the Critter Solutions is surveillance/pest monitoring. The trouble with using conventional cameras for detecting pests is that the quality ones can be costly and batteries are quickly drained. A bigger problem is a skewed or inaccurate picture of the seriousness of a pest incursion because of aversion behaviour – the light beams and clicks and whirs of the devices scare off easily-spooked pests.
“And there are so many false triggers. Anyone in my team constantly groans when we run our big camera trials because we’ll get tens of thousands of false images from wind-blown grass or a bit of fern frond. It fills up the SD card really fast and it takes hours and hours to manually sort through the images.”
Critter Solutions is developing a low cost, low power camera device that integrates all the AI classifications they’ve developed over the last few years.
“So you’ll get real time notification of your pests to the species level and that will all be displayed on a visual dashboard. So basically you can look at all your results online, or just the species you’re interested in,” Dr Blackie said.
The cameras and traps don’t need to be in a cellphone coverage area as there is a multi-communications solution available that also integrates with satellite.