What if the future of seed crop certification doesn’t walk the rows—but flies above them?
At InnoTech Alberta, Jan Slaski is piloting a bold new approach that uses machine vision and artificial intelligence to automate the most labor-intensive part of seed inspection — starting with industrial hemp.
Despite being federally legalized in Canada over 25 years ago, industrial hemp still fights an uphill battle against cultural misunderstandings. Its low tetrahydrocannabinol (THC) content — under 0.3% — renders it incapable of producing psychoactive effects, and yet many farmers remain hesitant to grow it. Slaski, originally from Poland where hemp has a long agricultural history, saw the opportunity early.
“I’ve been working with hemp for 23 years,” he says. “It’s the only true multi-purpose crop I know — grain, fibre, cannabinoids, and potentially livestock feed. It can play a critical role in sustainability and carbon capture.”
But certifying seed fields for hemp — particularly monoecious varieties, which contain both male and female flowers on the same plant — poses a significant logistical challenge.
When Flying Toys Meet Field Science
Traditional field inspections require human assessors to visually confirm plant sex ratios, an especially cumbersome process for hemp due to its towering stalks and uneven canopy. Inspectors often struggle to navigate the dense foliage and sample enough plants for statistically meaningful results.
Slaski’s solution? Let drones do the legwork — and AI do the thinking.
Using commercial-grade DJI drones equipped with high-resolution cameras, his team flies over experimental hemp fields, capturing thousands of overlapping images. These aren’t simply pretty pictures for crop health snapshots. They’re raw data, fed into deep learning models originally designed to inspect composite wood panels in construction.
That’s right: Slaski drew inspiration for hemp canopy analysis from software designed to identify fibres in oriented strand board (OSB). “If we can count every fibre in a messy board, why not every plant in a messy field?” he reasoned.
From Blueberry Muffins to Chihuahuas
In a whimsical metaphor that encapsulates the challenge of visual recognition, Slaski points to a meme showing what looks like a tray of blueberry muffins — until, on closer inspection, it turns out to be photos of muffins mixed in with Chihuahuas. The joke illustrates the stakes of false positives in automated crop inspections. It’s not enough to collect images — you need to train your model to know what it’s seeing.
And so the real innovation isn’t just the drone flights. It’s the multi-step pipeline that follows:
- Orthophoto correction to fix geometric distortion from camera angles,
- Segmentation models trained to detect male and female plant characteristics,
- Ground truthing with on-the-ground manual counts to validate AI predictions,
- And finally, statistical outputs that inspectors can use to certify or reject a seed crop with far greater accuracy and efficiency than traditional sampling methods.
What once took hours of laborious walking and manual tallying can now be achieved in minutes — from the air.
From Canola Pathogens to Escaped Hemp
The technology’s implications go far beyond hemp. Slaski’s team is simultaneously applying it to disease resistance screening in canola, particularly for verticillium wilt. A gantry-mounted camera system in a greenhouse mimics human rating scales to identify plant susceptibility, sometimes even outperforming trained technicians.
And there are regulatory implications, too. Since hemp is a licensed crop, the unintentional spread of seeds into adjacent fields — via birds, wind, or runoff— can result in “escaped plants” that pose legal risks. Using the same drone-AI pipeline, these volunteer plants can be detected, mapped, and removed before they trigger enforcement actions.
Slaski is quick to emphasize that hemp is just a case study. The tools being developed at InnoTech Alberta could be scaled to other seed crops where certification, trait verification, or disease monitoring is required.
“We’re not just counting hemp plants,” he says. “We’re building the digital toolbox for the future of crop certification.”
And that future, he argues, is not 20 years away — it’s already buzzing quietly above our fields, camera in tow, collecting images of plants most farmers can barely see from the ground.
As the seed industry moves toward greater demands for traceability, sustainability, and regulatory compliance, this fusion of agriculture and AI may become more than a novelty. It may become the new normal.
—This article is based on a live presentation by Jan Slaski of InnoTech Alberta, delivered at today’s meeting of the Canadian Seed Growers’ Association in Victoria, B.C.
