Researchers from the University of Illinois Grainger College of Engineering created a method that could allow farmers to predict their farm’s health in the same way that weather channels predict the weather.
Crop intelligence via AGMRI solution for growers, agronomists, ag retailers, and other players in the ag ecosystem. Source: Intelinair
The team gathered a dataset, called Agriculture-Vision, of large-scale aerial images of farms. Agriculture-Vision allows an algorithm to analyze patterns in areal images of farms, giving farmers actionable insights into crop performance. The goal is to help farmers improve their yields and make better decisions. The team created a computer vision technique that helps solve complex pattern-recognition problems with deep learning methods.
The current Agriculture-Vision dataset includes close to 100,000 images from thousands of corn and soybean fields across the Midwest. This includes annotations for conditions like nutrient deficiencies, drydown, weed clusters and more.
Agriculture-related visual pattern recognition has progressed slowly, partly due to a lack of large scale and high-quality datasets. This is a challenge because it requires recognition of patterns that do not occur consistently. It has been difficult to create a lot of high-quality agricultural image datasets because the image size is large and it can be difficult to recognize patterns that don’t occur consistently.
The team plans to expand the dataset to include different modalities. If images are captured season after season, it would allow deep learning models to predict the future of a farm and help farmers plan for long term farm health.
A paper on this technology will be presented at the IEEE/CUF Conference on Computer Vision and Pattern Recognition.
