New Computer Model for Early Detection of Biosecurity Threats

Paul Benden, a master's student at the University of Canterbury, is currently six months into developing a computer model that can identify biosecurity threats before they become a wider issue.

The work, funded by Biological Heritage National Science Challenge, uses regular photo updates to create a digital twin of the tree. The computer programme compares the digital twin to the regular photos it receives and alerts the handler if the tree is not growing healthily.

For instance, if urban trees in Auckland were infected with a new disease at an increasing rate, it could be expected that over time, the disease would move down the country and eventually hit Dunedin because of the lack of ability to detect diseases early.

However, with the computer model, we would be able to detect early stages of the infection in Auckland by noting the differences between the tree and its digital twin, giving us time to stop the disease in its track.

“Identifying biosecurity threats in urban areas is a tricky and time-consuming task, even for skilled humans,” says Paul.

“The computer model is designed to identify threats soon after they reach urban air and seaports, preventing them from spreading to the country's native forests, saving time and resources.”

The project supervisors, Dr Varvara Vetrova, Associate Professor Steve Pawson and Professor Richard Green, approached Paul with the concept. Now, six months into the project, Paul is successfully training the computer model to segment trees, which is no easy task.

“One of the big obstacles of this project is the uniqueness of each tree. Segmentation is mostly used for fixed items with straight lines – houses, roads, cars, so I knew that this would require a lot more skill than usual,” says Paul.

“However, the segmentation process has proven to be very successful. Each of the tests we have done has correctly identified different trees, even when they are overlapping one another.”

Paul presented early work on the project at the TAIAO Workshop and received valuable feedback and encouragement from members and recently presented his computer model at the fourth International Congress on Biological Invasions, which was well received.

Paul is currently implementing suggestions from the TAIAO Workshop to improve the model's performance.