VIDEO:
A combination of artificial intelligence and scientific ingenuity looks set to be the next step forward in protecting
Aotearoa New Zealand’s lakes and rivers from invasive aquatic weeds.
Management and detection of invasive submerged weeds costs millions of dollars annually, but NIWA researchers have
developed a way to detect and identify submerged weeds. This technology will enable agencies to survey far larger areas
more efficiently than is currently possible, and potentially lead much faster responses to new incursions.
Invasive submerged weeds can degrade water quality, exacerbate silt and flooding, reduce the number of native animals
and plants and play havoc with irrigation water delivery and hydroelectric power schemes
NIWA has developed a portable invasive species detector module that can be strapped to survey boats. The prototype is
housed in a small waterproof case with an underwater video camera attached. Inside is a computer containing an
artificial intelligence-based detector that has been trained to identify targeted invasive weed species and log their
locations in real time.
Principal technician Jeremy Bulleid has implemented a deep learning neural network - an artificial intelligence function
- to train a computer model to recognise two of New Zealand’s worst invasive weeds - lagarosiphon and hornwort - and
record their GPS locations. These data can then be exported to a mapping programme to enable control or eradication
strategies to be implemented.
"The deep learning process enables us to replace the human eyes and brain with a video camera and a computer by running
a detection application that has been taught what to look for," Mr Bulleid says.
Training a detector requires significant computing power and, depending on the complexity of the search environment and
‘target species’, may take days or even weeks. However, once training is completed, the trained detector is efficient
and can be embedded into the computer located inside the detector module for real time detection.
NIWA has successfully processed video imagery captured from an autonomous boat in a flume facility in Hamilton planted
out with three different submerged plant species.
The research is still in its early days and requires further fieldwork, data collection and software development to
evaluate its true potential.
However, NIWA freshwater ecologist Dr Daniel Clements says early detection and prevention is critical for achieving
effective freshwater biosecurity outcomes.
"If you can detect high risk invasive species early, before they are widespread, and implement effective management
strategies, you minimise the long-term economic, environmental, social, recreational, and cultural impacts caused by
these species.
"The development of these detector modules will enable a rapid and cost-effective detection and mapping that can be used
over large areas."
Currently most invasive species surveillance work is carried out by specialist divers. Dr Clements says the new
technology has the potential to shift diver expertise from detection effort to implementing control strategies.
"Real gains could be made by operating the modules from fully autonomous surface vessels that can be programmed and
deployed without constant supervision.
"Eradicating a freshwater invasive weed by detecting it early is much more feasible and cost-effective than dealing with
a widespread incursion in the long term."