James Cook University will oversee the development of a smart weed spraying robot that could reduce herbicide usage on sugarcane farms in Great Barrier Reef catchment areas by at least 80 per cent.
The two-year project, funded by a $400,000 grant through the partnership between the Great Barrier Reef Foundation and the Australian Government’s Reef Trust, is a collaboration between JCU, AutoWeed, and Sugar Research Australia.
AutoWeed is a startup agricultural technology company developing smart spot spraying systems, and Sugar Research Australia will help assess the water quality improvements the new technology promises.
JCU senior engineering lecturer Dr Mostafa Rahimi Azghadi will lead the project. He says the group’s effort will help reduce herbicide runoff, which is a serious threat to plants and animals in rivers, creeks, and coastal and inshore areas.
“Most herbicides, being mobile in soil, are carried in river runoff and have been detected in Great Barrier Reef ecosystems at concentrations high enough to affect organisms. Sugarcane farms are only 1.4 per cent of the catchment area but contribute 95 per cent of the pesticide load draining to the Great Barrier Reef ecosystems,” Azghadi says.
The project will rely on the pioneering deep learning technology being developed by JCU and AutoWeed to detect and spray priority sugarcane weeds.
Image based learning
AutoWeed co-founder and engineer Jake Wood says the new system will use stored images of weeds to detect and spray them without hitting non-target crops.
“Extending our AutoWeed spot spraying technology to sugarcane requires significant new research and development. We aim to reduce knockdown herbicide usage on sugarcane farms by at least 80 per cent.”
“This will incentivise water quality improvements in reef catchment areas by reducing weed management costs for farmers while also lowering the concentration of herbicides in runoff to support a healthy reef.”
In the first year of the project, hundreds of thousands of images of sugarcane farmers’ crops will be collected, labelled by a human expert, and fed into deep learning models to train the weed and crop detection system.
Every time the spraying system is used it will collect more data, so the deep learning models can further improve their performance over time.