Machine learning helps Manitoba farmers fight one of the most serious threats to wheat production.
Christopher Henry, an associate professor at U of M's Department of Computer Science, is a major project to apply artificial intelligence (AI) to agriculture. His team is developing automated tools to detect and measure Fusarium head blight, a fungal disease that damages crops across the prairie.
“We are working on developing automated methods to quantify the extent of fusarium head withering outdoors,” Henry said. “If we breed fusarium head depletion-resistant varieties, so that it helps breeding programs, […] Produce seeds […] It is resistant to this particular disease. ”
One important part of the research project is the 6-foot square “Data Rover.” Remotely controlled carts move fields and collect image and sensor data that can be used to train machine learning models. Two prototypes have already been tested, Henry added.
Beyond field robots, his lab also tackles machine learning theory. One focus is the use of generation methods. It's the same type of model behind large language models such as ChatGpt and Gemini. The task is to create a labeled image data set. Such datasets are scarce in agriculture, and invasive weeds and crop disease photos are less readily available than everyday images seen online.
“How is one thing we see [we can] Use some of these generation methods that are really good to create photorealistic images […] Henry pointed out. Using these datasets, the goal is to solve important problems to “train another machine learning algorithm” and to build digital tools useful in the agricultural industry.
Henry's interest in digital agriculture began in 2015 when he collaborated with a local robotics company on autonomous agricultural equipment. The project revealed that while the algorithm was present, the required labeled data was not present. The gap led him to develop new approaches to data creation and machine learning for agriculture.
Today, Henry works with industry partners such as Macdon and Manitoba Hydro, from machine learning and generative methods to hydrological engineering. He believes his work contributes in two ways. Advance machine learning theory and apply these methods to solve real-world problems.
“These methods of generating labeled datasets are very exciting because there are a lot of scientific areas, not just agriculture where labeled datasets are needed for highly specialized scientific questions,” emphasized Henry. “But creating a dataset is a costly and time-based part of that whole process, so we're excited to develop these methods and alleviate that bottleneck.”
For students interested in this field, Henry emphasized the importance of interdisciplinary collaboration. In digital agriculture, he explained that computer scientists are currently working with people in plant science, soil science and biology.
Henry said, “It requires people with knowledge from a variety of specific fields to come together to get together to get real-world solutions that help people.”
Ultimately, his goal is to create an OR […] Help us promote sustainable strengthening of grassland crop production. “In other words, it automates tasks with AI to maintain and even increase crop production while reducing dependence on fertilizers, pesticides and fuels.
