TO help growers pinpoint weeds in-paddock for site-specific control, the SA Grain Industry Trust is funding research into an innovative new weed identification and mapping system called the H Sensor.
The system involves new European technology that distinguishes different weeds in crops.
It is the first time research into the sensor has occurred outside of Europe.
Researcher Sam Trengove, conducting the project on behalf of SPAA Precision Agriculture Australia, says the system will give growers the tools to adopt site-specific weed management, resulting in more efficient use of herbicides, reducing their use, while giving the desired weed control.
"The goal is to hone the sensor technology to SA conditions so that it can pick out differences between our weeds and crops such as, for example, brome grass in canola or bifora in wheat," he said.
"The unit could be mounted on boomsprays for the sensor to capture images of weeds in-crop. There is potential to use those images for on-the-go application to turn the boom on and off as needed, however this will depend on the weed control treatment decision.
"Alternatively, it could be mounted on a tractor performing another paddock operation or on a robot that works autonomously."
The sensor, which costs $20,000, has a series of LED lights for illuminating the sensor's field of view.
The light source allows the sensor to operate regardless of light conditions, whether it is bright sun, overcast or at night.
It collects two images in the near infra-red and red regions of the light spectrum. Using these images, the sensor can detect all green plant parts against a background of soil and crop residue.
Crop plants are then differentiated from weeds based on leaf and plant-shape parameters.
The density and location of identified weeds can then be logged with a GPS reference and mapped.
The sensor is being adapted to SA conditions to identify weeds in lentils, lupins, wheat, chickpeas, field peas, canola, barley and faba beans.
In Europe, sensor development has focused mainly on wheat, canola, maize and sugar beet.
To identify weeds in Australian crops and conditions, new classification databases must be developed.
This requires collecting images from the field which represent the crop and weeds, manually classifying the weeds and crop in the image and then using this to build classifiers that will run in the sensor.
"This requires working with technicians from Agricon in Germany with access to their expertise and software to build new classifiers for each crop," Mr Trengove said.
"German technicians will visit SA in February to provide more training on the sensor and associated software. The sensor can only see and identify plants in direct line of site. Weeds hidden under stubble in no-till cropping systems may prove difficult to identify, so trials will be established to ascertain the importance of this."