They’re starting with Potato Virus Y (PVY), but they won’t stop there.

A team of Canadian scientists is working with a full range of regional industry stakeholders to commercialize an AI system called AgriScout for early identification and mapping of PVY infections – and eventually the early detection of other pests and diseases in potatoes and beyond.

AgriScout is the brainchild of Gurjit Randhawa, assistant professor in the School of Computer Science at the University of Guelph, and Aitazaz Farooque, professor of precision agriculture at the Canadian Centre for Climate Change and Adaptation, University of P.E.I. (UPEI). A close collaborator, Dr. Mathuresh Singh, recently retired from serving as the managing director at a renowned regional lab called Agricultural Certification Services (ACS) in New Brunswick. Other close collaborators include staff at P.E.I. Potato Board (especially Research & Agronomy Specialist Ryan Barrett) and Potatoes New Brunswick, along with several very experienced potato growers.

PVY is one of the most economically-significant diseases for Canadian potato farmers – and it’s especially challenging for seed growers, who are required to keep PVY levels under a very low threshold. Dr. Khalil Al-Mughrabi, potato pathologist of the NB Department of Agriculture, Aquaculture & Fisheries, notes that the timing of PVY detection is very important for effective management of this disease in seed potatoes, and the automated AgriScout system therefore offers a suitable solution to quickly identify PVY-infected potato plants and prevent the spread of the virus.

Indeed, Randhawa explains that right now, it’s next to impossible for seed growers to adequately scout for PVY. “It requires manual examination by very experienced scouts who must check the leaves of each and every plant for symptoms,” he says. “This takes a long time, and there is a severe shortage of experienced scouts. Identifying this disease is not an easy task for anyone.”

Farooque adds that “with the effects of climate change, I think we will see more viral strains emerging. We don’t know what the future will hold for PVY, but it will likely be more challenging with varying symptoms and strains.”

Origins of the system
The project began a few years ago when the PEI Potato Board and Potato NB reached out to Randhawa and Farooque about the challenges of dealing with PVY amidst a huge lack of experienced scouts.

Other technology firms and academic groups around the world were starting to work on machine vision and AI systems for detecting crop diseases and pests, so the Randhawa and Farooque tested the feasibility of using this tech, starting with some images from potatoes grown in the greenhouse. “We developed our own robot,” recalls Farooque, “and then added the machine vision system onto it, capturing high-resolution images along with the GPS coordinates of where the image is taken, known as ‘geo-tagged’ images.” As these images are taken, AgriScout’s AI analyzes them for signs of PVY infection and then directs the farmer to specific plants for treatment or removal, sharply minimizing the effect of this devasting disease.

“We wanted to be able to develop an interactive map at the end of the day that will guide the grower automatically to specific areas that are infected so those plants can be pulled as quickly as possible,” says Farooque. “You will have the map on your phone and are directed to the right location.”

Barrett notes that PVY reduction has always been a priority for their seed improvement working group, and growers have repeatedly noted the lack of trained roguers available. He calls the AgriScout work so far “very impressive” and says “we’re excited about the potential to bring this technology to commercialization in the near future.”

Training AI
Because scouting for this disease is not easy for humans, it’s also not easy for an AI. “There are many different potato varieties combined with different strains of the virus, in addition to different soils and climates in Canada, so you will end up with different leaf symptoms,” Farooque says. “You’re not going to necessarily see the same symptom of infection every time, which makes it really challenging for both human scouts and AI models.”

Another challenge too, noted in the team’s first year of developing the system, was sun angle. “If the sun, and thus the plant leaves, are too bright and we are capturing images with a great deal of light reflection, it’s all the harder to identify symptoms,” Farooque explains. “So, based on that experience, we added a shading mechanism to provide consistent imaging, so that the patterns and symptoms of the virus are much more visible, making it easier for our model to pick out those infected plants. And your model also has to be given images taken on cloudy days, on rainy days, early in the morning, late in the evening, every time of day and night and weather condition. The more variables you have in your data, more foolproof your model is going to be.”

Randhawa also explains that developing the system in collaboration with farmers was very important because it then has a much higher chance of widespread adoption. “They have seen firsthand all our research, testing and validating the images from their commercial farms,” he says, “Because it’s been developed with them all along, I think there is quite a bit of appetite from the potato growers for this system.”

Full collaboration
In the process of training the AI to recognize diseased leaves, working with Singh (and yler MacKenzie, R&D coordinator) at ACS was invaluable. “Dr. Singh is known as a guru of PVY, so we approached him and he was pleased to be involved,” says Farooque. “His lab is the only one in Atlantic Canada, I believe, where there is analysis of fields on a regional scope to determine where PVY is occurring, patterns of infection and so on. So that’s where we did our testing and trials. From a pathology perspective, we inoculated PVY at very low levels – one, two, three per cent infection – to build a database of images to train the AI, with all the possible symptoms and different variations in the symptoms at different infection levels.”

Working with specialists like Barrett was also invaluable. “We needed the model to understand when symptoms are not PVY, they are symptoms for something else, and we’ve had a very close interaction and collaboration with Ryan and other team members to figure out what we are collecting in terms of images,” says Farooque, “and to make sure we are identifying everything correctly for the model to train on.”

MacKenzie explains that ACS Lab led a pilot field validation of AgriScout, using lab diagnostics as the basis for validation, complementing the extensive work of Randhawa and Farooque in comparing AgriScout and human visual scouting. “We followed the robot through a large, PVY-infected Russet Burbank field, collecting ~500 leaf samples from imaged plants for lab confirmation of PVY status,” he reports. “This rigorous validation achieved an overall accuracy of 76 per cent accuracy in AgriScout correctly identifying PVY-infected plants and rejecting non-infected plants, with a particular strength in avoiding false positive detections. These results give us confidence in the AgriScout’s potential as a decision-support tool for PVY management.”

AgriScout is already on par with human scouts. Over the next two years, the team will further test and refine the technology with farmers and ACS Lab, with commercialization to follow.

Keeping farm costs low
As AgriScout moves towards the marketplace, there will also be a focus on ensuring it’s as low-cost and convenient as possible. “We don’t want to increase farm expenses,” says Randhawa. “We want the system to be integrated easily onto a tractor, sprayer, cart or ATV, whatever piece of equipment farmers already have and works best.” AgriScout is also based on simple parts that are easily integrated, note Randhawa and Farooque, which they think will also make its cost significantly lower than competitors.

Taking a look at ROI for farmers, if AgriScout is set up on a sprayer or tractor that’s legally permitted to operate autonomously, farmers can therefore run AgriScout at night and use the same piece of equipment for other things during the day. “That’s where part of the return comes in,” says Randhawa, “but also through improved disease control. We did a rough estimate for an average potato farm, and we have found ROI is achieved quickly, but we will be performing a detailed economic analysis. And the bigger the farm size, the faster the ROI.”

Most of all, this team is very excited to be solving an impossible problem for potato farmers. “The shortage of scouts really is severe,” says Randhawa. “You cannot just send anyone to go and look for those diseased plants. You need someone which has decades of experience who can look by eye on those plant leaves and then say with surety, this is infected. Scouts also need good weather conditions to go into the field, and they can cover only about three to five acres a day. But now we will be making available a system that works 24 hours a day and covers a huge area quickly and is within reach econ...

Potatoes in Canada

Message History

Delivery option(s)