Developing prairie-centred machine learning algorithms
EMILI is working with the TerraByte research team at the University of Winnipeg to develop a publicly accessible database of labelled images of plants and weeds which will be used to train machine learning models for Prairie-centric plant phenotyping, disease assessment, and weed management. This will allow us to develop and test crop scanning technology capable of identifying weeds amongst Western Canadian crops.
These data sets are central to developing digital agriculture solutions that work in Manitoba. Breeders can use the phenotyping data to help select for more tolerant, higher yielding crops. Farmers will use weed identification Al to identify, measure, locate and assess the severity of weeds or diseases in their fields, allowing them to make site specific data driven decisions.
Testing and validating technology solutions
Collecting images of prairie crops and weeds
EMILI, in collaboration with the University of Winnipeg, is collecting imagery of field peas, soybeans, corn, canola, and hard red spring wheat at Innovation Farms. Millions of labelled images are produced in the lab and will be used to train machine learning models for plant phenotyping, disease assessment and weed management. This includes a robotic camera system for indoor imaging, a growth chamber, and outdoor field imaging at Innovation Farms.
Hyperspectral imaging of prairie crops and weeds
We are using hyperspectral imaging to collect data for machine learning models. This includes scanning different plant species (classification), multiple cultivars of a given species (phenotyping), and a single species under different stressors (disease detection).
Photogrammetry for plant research and breeding
We are developing low-cost photogrammetry systems specific for plant research and breeding. The 3D models produced by this technique provide a more accurate, robust and time-efficient means to characterize and track plant features.
Data Rover for in-field imaging
UWinnipeg's TerraByte research team is working with R-Tech Industries to develop data rovers for in-field imaging and assessment at the plant level. Large volumes of high-quality, in-field data are crucial for the development of crop scouting technology and to expedite plant breeding and agronomic research.
Using machine learning to assess plant health
The University of Winnipeg’s TerraByte project uses data collected at Innovation Farms for the development of a machine learning model for plant image classification. Using both indoor and field data, this model efficiently and precisely identifies weeds and disease within Western Canadian crops. The project also integrates 3D multispectral scanning, 3D photogrammetry, and hyperspectral imaging. These technologies aid in detection and identification of weeds, diseases, and give a better assessment of overall plant health. They can detect subtle differences between plants that are imperceptible to the naked eye.
Developing a public plant image database
By making the data we are collecting accessible through the Digital Research Alliance of Canada, researchers and others working in digital agriculture can access the information they need to accelerate the speed of innovation. If you are interested in learning more about how to access this data, email email@example.com
Learn more about machine learning to grow digital agriculture
Watch this 2020 video to learn more about the ways advancements in technology, autonomous robotics, and machine learning are revolutionizing the agriculture industry. Intelligent technologies produce vast amounts of data to drive insights, decision-making, research and other value-added components. Insights derived from this data will increase crop yields, expedite crop breeding and optimize input application and other agricultural practices.