Name that pest – Using AI to rapidly ID pest species

Name that pest – Using AI to rapidly ID pest species and biosecurity risks

Ever been fined $400 at the border for a forgotten banana and wondered about the strict biosecurity rules in New Zealand? A University of Canterbury academic is working to make it possible to instantly identify suspect insects, plants and fungi from a photo using artificial intelligence (AI) and big data.

University of Canterbury lecturer Dr Varvara Vetrova, in UC’s School of Mathematics and Statistics, is leading the MBIE-funded, three-year $1 million project, with Manaaki Whenua (Landcare Research) and University of Waikato colleagues, researching AI use, such as machine learning*, in protecting the future of New Zealand agribusiness and biosecurity.

The project – ‘Biosecure-ID’: Machine learning to automate image-based identification of species – will focus on developing and testing new approaches for identifying biological species from photos, based on deep convolutional neural networks.

“We are developing a prototype image-based system for rapid, automatic identification of potential pest species in New Zealand,” Dr Vetrova says.

“Provided that we have a snapshot of an organism, we could tell which species it is. So a biosecurity officer or a farmer could find a bug, fungus or plant, take a picture and from that the programme can quickly tell them whether this is a worry or this is a New Zealand species.”

The researchers are working to develop an app that could tell what a specimen is from a photo. Plants, fungi and insects would be automatically identified, with similar-looking specimens easily distinguished from almost identical but separate species.

“We are trying to build a targeted model, the main idea is that an interested group, for example farmers, will ask to help identify particular weeds. Then they will supply us with few images and we will train and provide a model for them,” she says.

“At this stage it is still a prototype and we are focusing on three very specific case studies – plants of the Coprosma genus, New Zealand moth and fungi groups. We are planning to add several other case studies and build a framework of expandable targeted models which are New Zealand specific.”

This project is a part of government-funded Endeavour grant and involves researchers from the fields of biology, computer science, computer vision and applied machine learning.

*Machine learning is a subset of artificial intelligence that often uses statistical techniques to give computers the ability to “learn” with data, without being explicitly programmed.

© Scoop Media