Accounting for about 15 per cent of flora nationwide, invasive weeds negatively impact crop productivity and lose Australia’s agricultural industry an estimated $4 billion annually, with their estimated cost to the natural environment even higher.
The scale of Australia’s weed problem is such that it cannot be managed by manual weed-spraying alone.
Which is why Alex Olsen, a PhD student at James Cook University, under the supervision of JCU’s Professor Peter Ridd, is developing a robotic system armed with multiple cameras connected to spray nozzles that can pinpoint, then target weeds with far greater speed and precision than conventional weed spraying equipment can.
The robot is being developed by student Alex Olsen as part of his doctorate in Engineering and Related Technologies.
Olsen’s goal is to create a robot that can differentiate undesirable weed species from other plants. To do this, he’s employing a sophisticated computer algorithm that enables the weed- bot to detect an array of key variables including plants’ colour, shape and texture.
The weed-killing robot’s camera ‘eyes’ are designed to simulate how an experienced farmer would differentiate weeds from desired plants, such as crops and native species.
“We try to think of it as what we do when we first see it with our own eyes,” Olsen explains.
“We can identify green straight away, then determine the shape of the plant, then the leaves, and then we strip that down to a point where we can grab small ‘texture windows’ for a feature comparison to find out if it matches what we’re looking for.”
The front of the cutting-edge weed-killing robot will be equipped with multiple cameras, with herbicide sprayers on the back, the two working in tandem through a series of computer-mediated processes that ascertain precisely where and when to spray for weeds, thus eradicating more weeds while eliminating overuse of herbicides.
The team trialled an early prototype of its precision weeding machine recently at a site in Hidden Valley in North Queensland’s Paluma Range; saying its positive performance in the field “is encouraging”.
Once the prototype has been tweaked and its technology refined, the new ag-bot will be used to target environmental weeds in the first instance, but has obvious broader applications in agriculture and species identification.
Olsen says the JCU team see the device being fitted to, or towed by existing farm vehicles in the short-term, and as a fully autonomous agricultural vehicle down the track.
And the new robot’s applications stretch far beyond invasive weed removal. “It could be for helping to identify which species of plant are located in a national park,” Olsen says.
“The algorithms are for image processing: if you have a knowledge of this, it can apply to anything.”
The JCU team’s work on weed-killing robots netted Olsen and Professor Ridd a runner-up award in the Digital Image Computing: Techniques and Applications (DICTA) category of Canon Australia’s Extreme Imaging competition.