Separating weeds from the wheat
by Rebecca Gill
Wheat
is Australia’s most widely grown and economically reliable crop, but its
associated grass weeds are a problem that farmers, agronomists and
researchers constantly struggle to manage.
However, computerised technology currently being developed by PhD student Mahmood Golzarian, under the primary supervision of Dr Sang-Heon Lee, Program Director of the School of Advanced Manufacturing and Mechanical Engineering, could potentially open new avenues for more effective and more efficient weed control.
The common weed control method in conservation cropping is to blanket spray entire fields with chemical herbicides, which results in significant chemical wastage affecting environmental sustainability and farm economic viability.
Alternative "patch spraying" techniques apply herbicides only to field areas above a given infestation level, using remote sensing technology to define these weedy areas. In a next step, researchers are now conceiving "spot spraying" techniques targeting small groups of weeds or even individual weeds. With this technique, accurate weed location information is combined with computer processing to apply optimum doses of herbicide.
Golzarian’s research, which uses advanced algorithms and computer vision technology, could significantly contribute to this weed control revolution by accurately distinguishing between similar looking weed and crop seedlings.
It is no simple matter, as many weeds often have comparable shape and colour features to growing crop plants.
"In wheat cropping, common weeds include brome grass, rye grass and wild oat. Among these, rye-grass seedlings are the easiest to distinguish from wheat seedlings because they are different enough in terms of colour and shape parameters," Golzarian says.
"But the morphology of wild oat is visually very close to the wheat plant, particularly in the early growth stages. So the most difficult task in developing this system is finding some features that are unique for each plant type."
Golzarian’s associate supervisor, Dr Jack Desbiolles from the Agricultural Machinery Research and Design Centre, says accurately classifying grass weeds within a wheat crop using digital image analysis techniques is a real challenge, in part due to non-uniform soil and residue backgrounds normally found in conservation cropping contexts.
"Mahmood’s project will provide the building blocks for future PhD
projects to advance this technique into a real-time computerised application
allowing spot treatments of weeds, and usable by farmers,"
Dr Desbiolles says.
"His research represents the first step to getting computers to do what farmers, agronomists and agricultural researchers currently do by eye when assessing weed and crop establishment in the field. Essentially, it is about automating a process which is time and resource intensive, and subject to human misjudgement.
"With an estimated $4 billion a year associated with weed control
practices and grain yield losses, the economic savings and environmental
benefits associated with making weed control more targeted, more efficient
and possibly non-chemical, are considerable."