Scientists at the Rochester Institute of Technology will use a grant from the New York State Department of Environmental Conversation (DEC) to identify and map five invasive plant species around New York State, the university said on 20 June.
Instead of using traditional methods of data collection, the two faculty members plan to train an artificial intelligence (AI) to recognise the plants in images from Google Street View and other sources, allowing officials to identify high-priority treatment sites, identifying where the species have spread and subsequently prioritising where to intervene.
Invasive plants can quickly damage ecosystems, cause economic hardships for farmers and pose health risks to the public, but regulators tasked with halting their progress often have limited resources.
The two researchers, Assistant Professor Christopher Kanan and Associate Professor Christy Tyler, believe that this new approach will be “a big improvement over current tracking methods, which rely on field reports and volunteer crowd-sourcing”.
“There are limitations to crowd-sourcing,” Kanan said in a statement. “Rural roads are hard to get to and hard to monitor and often you can’t even pull over to the side of the road easily.”
“If we can figure out where these species are, using artificial intelligence, organizations like the Department of Environmental Conservation can better allocate their resources toward managing these plants,” he added.
Kanan and Tyler have already identified working models for two species – common reed (Phragmites) and Japanese knotweed—which they will further fine-tune, the university said. They will also develop new models for identifying giant hogweed, tree-of-heaven and purple loosestrife.
These plants are some of the most problematic species in New York because they quickly crowd out native plants, are hazardous to humans or host insects that wreak havoc on crops.
“We started with two target species that are pretty obvious, noxious, public enemy number one-type species that are common on roadways,” Tyler explained. “We picked them because they are priority species and distinctive looking plants that can be picked up easily by computers.”
“Moving forward, tree-of-heaven is an important one because it’s the host of the spotted lantern fly, a new invasive insect that has huge potential for crop damage, especially tree fruit including apples, pears, peaches and grapes. That will be a bigger challenge because it looks a lot like other species,” she said.
Once the computer models are functional and begin making predications about where the invasive species are located, Tyler’s environmental science students can verify the identification in the images and in the field to confirm that the algorithm works.
Students in the environmental science senior capstone class, with Visiting Assistant Professor Kaitlin Stack Whitney, have already helped with this verification process. Ultimately, the results will be communicated to managers through the New York State Partnership for Invasive Species Management (PRISM).
“Most of the big successes in computer vision use deep neural networks with images that are fairly small. Google Street View images have over 1,000 times more pixels . . . images typically used. Standard algorithms totally break down in this situation,” the researchers said. “
We have developed efficient methods for detecting regions of these high-resolution images that are likely to contain plant matter and then we determine if an invasive plant is present in the scene,” they concluded. “For needle in the haystack type problems, this multi-stage process seems to be very successful.”
The project is funded by the DEC’s Invasive Species Grant Program, which is designed to “support projects that target both aquatic and terrestrial invasive species”. The DEC received 96 applications and awarded approximately US$2.8 million from the New York State Environmental Protection Fund to 42 projects.