It can tell the difference between a wallaby and other mammals, and will eventually be able to send a text or email message to its base, advising of the sighting.
Otago Regional Council, in association with Environment Canterbury, set up the $25,000 project, and put a trained camera on Aviemore Dam late last year as a trial.
Once the trial is complete, they will install it and others on sites in the region to better identify location and wallaby numbers.
Wallabies are considered a major threat to the region as they cause significant damage; three can eat as much grass as a stock unit.
ORC environmental officer Simon Stevenson has been running the council-funded trial.
He approached Christchurch technology developer the Cacophony Project, as their AI technology could already identify thermal images of other pest mammals, from cats and mustelids, to possums, rats, deer and pigs.
"I contacted them to see if they could add wallabies to their list [of identifiable animals] and they jumped on board."
A camera was installed in the Willowbank Wildlife Reserve’s wallaby enclosure in Christchurch, where the AI was taught to identify the heat image of the pest.
"It learned what the heat signature of the animal is when it moves and at different angles."
It took thousands of videos of the animals, and eventually learned to recognise them.
Mr Stevenson and Cacophony technicians had to look at each of the videos and manually classify the ones containing a wallaby.
It has 94% accuracy and is far more efficient at identifying the images compared to trail cameras that are triggered by movement.
"It is a great initiative for pest control in New Zealand."
