Open-Source Image Hub Enhances AI in Farming

Published Date : 9/10/2025 

A new open-source plant image repository developed at NC State University aims to accelerate the use of artificial intelligence in solving agricultural challenges. 

A major open-source image repository to be released nationwide this fall could be a significant step forward in helping unlock artificial intelligence’s potential for solving stubborn agricultural challenges.

Led by the U.S. Department of Agriculture’s Agricultural Research Service and NC State University, the Ag Image Repository, or AgIR, is a growing collection of 1.5 million high-quality photographs of plants and associated data collected at different stages of growth.

The collection will first be released on the high-performance computing cluster SCINet — a first step toward making the resource freely available worldwide to agricultural researchers in both public and private sectors.

Access to this data will be game-changing for plant intelligence technology around the world.

Meanwhile, the team is busy using those images to create what it calls “cut-outs” — plants removed from their background — that will be key to AI development. These include 16 cover crop species, 38 weed species, and a growing number of cash crops, such as corn, soybeans, and cotton.

Alexander Allen, head of the AgIR’s system software development team, says that the repository is designed to help researchers who want to create AI-based solutions for farmers, plant breeders, and others in agriculture.

It grew out of research by Precision Sustainable Agriculture (PSA), a nationwide network of scientists, engineers, farmers, and others with a shared interest in helping farmers make real-time, data-driven management decisions.

“The lack of publicly available, high-quality agricultural imagery has been a barrier to advancing machine learning research in agriculture,” Allen says. “Access to this data will be game-changing for plant intelligence technology around the world, leading to an increase in design technologies for precision agriculture to help farmers get the most out of their fields while protecting their fields and nature from ecological damage.”

AgIR could be especially helpful for those looking to develop agricultural tools and technologies that employ computer vision, a form of AI that can help machines “see,” understand, and respond to the world around them.

Computer vision is widely used these days — it stops driverless taxis from running over pedestrians, surgical robots from making the wrong cuts, and impostors from getting through airport checkpoints. Farmers, too, are using computer vision, most notably with tractor-mounted machinery that identifies weeds and sprays them with herbicides.

But for agriculture, arriving at effective computer vision solutions can be more challenging to develop than it is for some other fields, AgIR project leaders say.

Compared with factory floors, farm fields are complex, highly variable environments that can be difficult to navigate. And subtle differences in the way a plant looks can have profound implications for how it should be tended.

As Allen puts it, “A stop sign looks the same on the East Coast as it does on the West. But that’s not always the case with a pea plant.

“Even with a single plant species, you have genetic varieties that have different visible traits and that respond very differently to environmental factors,” he adds. “Not only that, a plant might look differently if it was grown in drier, hotter weather than if it was grown in cooler weather, if it was water stressed, and so on.”

To train AI models robust enough to account for the variability in farm fields and the plants that grow there, agricultural researchers need lots of carefully curated and annotated images depicting plants under different growing conditions and at different growth stages, Allen explains.

Computational agronomist Matthew Kutugata sees the AgIR as an important step forward in meeting that need.

“Agriculture doesn’t have the big, well-labeled image sets other fields take for granted. AgIR closes that gap so we can train models that hold up across farms, seasons, and applications,” says Kutugata, who leads PSA’s data engineering and computer vision efforts. “Because the data and baselines will be open, researchers, their students, small labs, and even growers have a clear on-ramp to build, test, and improve tools without starting from scratch.”

Kutugata, Allen, and NC State agronomist Chris Reberg-Horton see AgIR as a step toward realizing the full benefits of precision farming.

In precision farming, producers deliver exactly what a plant needs, precisely when and where and in the amounts needed. For example, rather than spraying an entire field for insects, which can be expensive, farmers could spray just the areas where they are a problem. That would protect crops, limit chemical use, and safeguard the environment from excess application.

Though precision farming has been talked about for over 40 years, “it has been aspirational for the most part,” says Reberg-Horton, a professor in the Department of Crop and Soil Sciences, director of the N.C. Plant Sciences Initiative’s Resilient Agricultural Systems Platform, and PSA’s co-director.

“Smart equipment is available now to apply most of our inputs variably, but we have been stuck on creating enough intelligence to tell that equipment what to do,” he says. “Computer vision is the technology that can do it.”

For years, Reberg-Horton has been developing computer vision systems that help farmers make the most of cover crops – ones grown mainly because they can enhance the value of future cash crops.

To start building the tools he wanted, he needed mountains of solid images of any plant – cash crop, cover crop, or weed – that might be encountered in agricultural fields.

He and his team soon realized that collecting images of individual plants in the field is tedious, time-consuming, and labor-intensive, and so is labeling the images with the data needed for building artificial intelligence.

So they built tools to streamline the process: robotic hardware to collect plant images that met exacting standards and software to make it easier to label the images with the data needed to train reliable AI models.

To automate the picture-taking, the team developed three robots that operate in a so-called “semi field” environment at three locations – a USDA research facility in Beltsville, Maryland, Texas A&M University, and outside NC State’s Plant Sciences Building.

Each of these wheel-mounted Benchbots carries a camera capable of capturing photos with the level of detail needed to make them usable for scientific research.

“We use a camera capable of capturing incredibly detailed images of plants unlike anybody has attempted before,” Allen says.

Each Benchbot is programmed to move across a “field” of hundreds of plants in pots that have been arranged in rows, allowing a camera that moves along an overhead track to take photos of each pot.

“If we’re imaging peas, for example, we’ll bring in a bunch of varieties of peas of the same species, plant them, and then every week we run at least three passes, so we end up with a time series as the plants grow and develop over the course of their lifetime,” Allen says.

While building the hardware, the team also developed software to help automate the process of cutting the plants out of images, making color corrections, and attaching detailed descriptions about each image.

“Our challenge has been not just collecting the images but making it reasonable for a human to annotate and provide the context so you end up with the size dataset that you need to train a machine learning model,” Allen says.

As they developed their hardware and software solutions, PSA team members realized that the database they’d assembled could be useful not just for their own research but for anyone hoping to develop AI-driven tools and technology for farmers.

Right now they are turning their attention to more than just farmers. Plant breeders can also make use of computer vision for high-throughput phenotyping, the automated, rapid, and precise measurement of an organism’s traits on a large scale.

“I’m excited about it being used for applications I never imagined, by teams I have never heard of, to impact farmer problems,” Reberg-Horton says. 

Frequently Asked Questions (FAQS): 

Q: What is the Ag Image Repository (AgIR)?

A: The Ag Image Repository (AgIR) is a major open-source image repository developed by the U.S. Department of Agriculture’s Agricultural Research Service and NC State University. It contains 1.5 million high-quality photographs of plants and associated data collected at different stages of growth, designed to help researchers create AI-based solutions for agriculture.

Q: How does the AgIR help in developing AI solutions for agriculture?

A: The AgIR provides a large, well-labeled dataset of plant images that can be used to train AI models. This helps researchers and developers create robust computer vision solutions that can identify and manage plants under various conditions, which is crucial for precision farming and other agricultural applications.

Q: What are some of the challenges in using computer vision in agriculture?

A: Agricultural environments are complex and highly variable, making it challenging to develop effective computer vision solutions. Subtle differences in plant appearance due to genetic and environmental factors require large, carefully curated datasets to train reliable AI models.

Q: What is precision farming, and how does AI play a role in it?

A: Precision farming involves delivering exactly what a plant needs, precisely when and where it is needed. AI, particularly computer vision, can help farmers identify and manage specific areas of a field, reducing waste and improving crop yields while protecting the environment.

Q: What are some of the applications of the AgIR beyond farming?

A: The AgIR can also benefit plant breeders by providing high-throughput phenotyping, which involves the automated, rapid, and precise measurement of an organism’s traits on a large scale. This can help reduce tedious tasks in the plant breeding process and accelerate the development of new plant varieties. 

More Related Topics :