Smart Ag 2020 — The best is yet to come

IntelinAir’s pioneering work in computer vision is about building a better agriculture for everyone.

AgMRI empowers growers and agronomists to make better decisions, and we’re making our product easier than ever to use with our new iPhone, iPad, and web applications that drive timely farmer actions for higher yields and profit.

All that is available today, but at IntelinAir, we’re also concerned about tomorrow, which is why we’re investing in the future of smart agriculture.

We’ve made publicly available our Agriculture-Vision database, a first-of-its-kind massive dataset of high-resolution aerial farm images that have been professionally annotated by top agronomists. You can find out more at

Why does this matter? Image recognition or computer vision is a significant driver of AI innovation. It’s what allows a self-driving car to “look” at the road ahead and avoid obstacles. It’s how machines can scan medical x-rays and other images to detect diseases in their early stages, often before a trained physician would spot them. The same technology can be a game-changer for agriculture.

A game-changer for ag

Imagery can spot not just disease, but any number of issues like weed escapes, insect pressure, and all other threats to yield. Having a system to interpret imagery will also be critical to regenerative ag practices that minimize the use of chemicals and conserve water.

These features are possible with deep learning algorithms that work by being “trained” on a set of labeled images. In the classic example, you have a stack of pictures of multiple breeds of dogs and a stack of pictures of cats of all varieties. The system will analyze each image and notice, for example, that if the animal in the photo has a long snout, it’s probably a dog. But some dogs, like pugs, have a very short snout—so the algorithm has to take many other features and measurements into account in a highly complex process.

The bigger the training set, the better. Say you had a small stack of photos to work with, maybe one lacking a pug. Your algorithm might mistakenly think the first one it sees is a cat. This is why our new dataset is—necessarily—huge.

Agriculture-Vision is massive


Our Agriculture-Vision database includes 94,986 high-quality aerial images from 3,432 farmlands across the US. Each image covers the normal visual spectrum as well as near-infrared. The photos are sharper than any other set of imagery that has come before it — each pixel covers up to 10 centimeters of the field. In other words, you can see everything in great detail, and it covers them throughout the season.

Critically, these aren’t just raw images. They’ve each been carefully annotated by skilled agronomists who have labeled each of the key features. So they’re just what a deep learning algorithm needs to learn what a healthy farm should look like, and where to look for problems.

Working with this incredible database is a challenge because the images are at such a high resolution, which requires a lot of processing power. AI techniques designed to work within the human visual spectrum also need to be modified to work with the multi-spectral imaging of our dataset.


Cloud Shadow

We’re just at the start of what’s possible in making the most of AI in agriculture. By putting in hard work now, we’ll see even more powerful AI systems that will improve agriculture for everyone.

So we’ve made a slimmed down 21,061 image version of Agriculture-Vision part of a challenge to help generate interest within the broader AI community to work on the future of agriculture. We’ll be part of the prestigious CVPR computer vision event, currently scheduled for June 16 in Seattle—as of now, it’s still going forward, but we’ll participate in whatever form it takes. We’re co-hosting a panel with great participants from CNH, Microsoft, IBM, and ADAMA to bring high-level attention to ag and AI. This is the kind of collaboration that will produce results.

Higher yield and greater profits

Computer vision is all about letting the growers and agronomists know in great detail exactly what’s happening in their fields throughout the season. Having aerial imagery and satellite imagery is not enough. A sparse cluster of shapes in a photo might not mean much to you, but a trained AI will pick up on the earliest signs of weeds. You’ll then know to go out and treat the area before the weeds become a problem.

More robust AI systems will be able to interpret the images so that the growers and agronomists have the insight they need to make better decisions. Smart agriculture means making fewer mistakes and enjoying higher yields—along with greater profits.

At IntelinAir, we’ve made smart agriculture a reality with AgMRI, but we know it can be even better. We have tons of new features in the pipeline, but there’s no limit to what else is possible with the best minds in AI working together on computer vision in agriculture.

You may also like