From a vision of AI in plant science to the reality of Pest-ID in global agriculture

AMES, Iowa - A 2016 research paper explored a vision for using emerging tools in machine learning – a branch of artificial intelligence – to help plant scientists study stress in plants.

 

Advancements in imaging technologies at the time “resulted in a deluge of high-resolution images and sensor data of plants,” wrote the paper’s four co-authors, all faculty members at Iowa State University who are working to turn their early vision for AI in agriculture to in-the-fields reality.

 

However – wrote Arti Singh, Baskar Ganapathysubramanian, Asheesh Singh and Soumik Sarkar, long-standing members of Iowa State’s “Soynomics” team that has studied how computing tools can improve agriculture – “extracting patterns and features from this large corpus of data requires the use of machine learning tools.”

 

Their paper in the scientific journal Trends in Plant Science went on to explain their ideas for using different machine learning tools for studying plant identification, classification, quantification and prediction.

 

The payoffs were obvious to the four Iowa State researchers: “One of the major advantages of using (machine learning) approaches … is the opportunity to search large datasets to discover patterns and govern discovery by simultaneously looking at a combination of factors instead of analyzing each feature (trait) individually.”

 

Yes, they wrote, “The outlook for (machine learning) tools in agriculture is very promising.”

 

Well, nine years later, artificial intelligence in agriculture isn’t just for the scientists. A tool developed by the four and many other collaborators, “Pest-ID,” is helping farmers and growers upload photos to identify and get recommendations for managing insects and weeds.

 

The tool was featured at Iowa State University’s booth at the 2025 Iowa State Fair under the headline, “Bugged?” Curious fairgoers asked if they could, really, as the fair display said, “instantly identify good, bad or just plain confusing insects.”

 

Well, yes, “With over 96% accuracy, the tool provides real-time classification and insight,” the display said, “into whether that insect on your crop is a pest, pollinator, or protector.”

 

A field trip to a corn field

After working to develop separate InsectNet and WeedNet applications, Iowa State’s AI-in-Ag team has combined them into one “Pest-ID” web-based tool.

 

“What would you like to identify,” reads the website. Users can click a button for an insect or for a weed. Then they’re asked to upload a photo of their pest.

 

Upload, say, a cell phone photo of a weed growing between corn plants in a field just north of Ames and, in just a few seconds, this is what you’ll see:

 

TAXONOMY

Scientific name: Amaranthus tuberculatus

 

IMPACT

Amaranthus tuberculatus, commonly known as waterhemp, is a highly competitive weed that significantly reduces crop yields. It competes aggressively for water, nutrients, and light, and can cause substantial economic losses in various crops, particularly in corn and soybean fields.

 

MISC.

Amaranthus tuberculatus, also known as tall waterhemp, can produce up to 1 million seeds per plant. This prolific seed production contributes to its rapid spread and persistence in agricultural fields, making it a challenging weed to control.

 

Want to learn more? You can type questions in the “Ask our AI” box:

 

“How do I control Amaranthus tuberculatus?”

 

“Control Amaranthus tuberculatus (waterhemp) through integrated weed management strategies including crop rotation, pre-emergence herbicides, post-emergence herbicides with multiple modes of action, and mechanical cultivation. Prevent seed production and spread by controlling plants before they set seed.”

 

The researchers said they’ve built this AI tool using expert-developed and -curated bulletins and papers, ensuring scientific rigor in the responses.

 

A one-stop ag AI shop

Arti Singh, an Iowa State associate professor of agronomy, said the next step in the development of Pest-ID is to expand the tool’s capabilities to the identification of plant diseases and advice for managing them.

 

“We want to build a one-stop shop for insects, weeds and diseases,” she said.

 

That’s not just important to farmers and the extension specialists who work with them. In today’s world of growing populations, extreme weather and crop pests destroying as much as 30% of food in the field, food security is becoming national security.

 

So how can farmers – and countries – defend their fields from invading pests?

 

“We thought AI could come to the rescue,” said Ganapathysubramanian, a mechanical engineer and Iowa State’s Joseph and Elizabeth Anderlik Professor in Engineering.

 

So, the research team began asking each other questions: Can AI really help? How can it help? What kind of data is needed?

 

The group started with images. First of insects. Then weeds. And now diseases.

 

The InsectNet tool, for example, is backed by a dataset of 12 million insect images, including many collected by citizen-scientists, according to a scientific paper published earlier this year.

 

Now, Arti Singh mentions datasets of 150 million images covering insects, weeds and diseases. Those images would cover crop problems around Iowa and all over the globe. The researchers have made sure the tools can be fine-tuned for fields and farms in different states or different countries.

 

Those “global-to-local” datasets are constantly updated.

 

“We know new insects, diseases and weeds will keep coming,” Arti Singh said. “There is a need for a robust, context-aware, decision-support system capable of early detection and accurate identification, followed by delivery of expert-validated, region-specific integrated pest management recommendations.”

 

The researchers are working with Iowa State Extension and Outreach specialists to share these and other AI-based tools with potential user groups, including 4-H clubs, FFA chapters and farmers.

 

Building their own ‘super-massive models’

That fine-tuning is made possible because Iowa State’s AI team has the expertise to make it work.

 

Sarkar, a professor of mechanical engineering and the director of Iowa State’s Translational AI Center, said the Iowa State team built what’s known as “foundation models” for the ag project.

 

“These are super-massive models that can be fine-tuned for different tasks,” he said.

 

They’re the kind of models usually built by big AI companies and not university researchers, Sarkar said.  

 

And what about the AI models we all know about? The ones assisting search engines these days? Couldn’t they identify a picture of an ag pest?

 

“Most can detect a charismatic species like a monarch butterfly,” Sarkar said. “But with specific invasive species, our models do much better than the general-purpose tools.”

 

Now government agencies are recognizing those models for their potential value beyond agriculture.

 

“We built these tools for the agricultural context to help farmers and extension specialists,” Sarkar said. “But now we’re receiving U.S. defense and national security interest for invasive species management or biosecurity. We’re seeing increasing traction from those communities.”

 

To cyber-agricultural systems

A February 2024 paper in the journal Trends in Plant Science, published eight years after their original review of AI in agriculture in the same journal, illustrates how far Iowa State’s team and its collaborators have advanced their ideas.

 

From scientists studying stress in plants using images and machine learning to identify patterns, the team has progressed to describe models of entire cyber-agricultural systems, including digital twins of crops, to inform decisions and enhance crop breeding and sustainable production.

 

“In a cyber-physical system,” the researchers wrote, “the physical space is the source of information, and the cyberspace uses the generated information to make decisions which are then implemented back into the physical space.”

 

Bringing cyber-physical modeling to agriculture “has enormous potential to enhance productivity, profitability, and resiliency while lowering the environmental footprint,” they wrote.

 

The research disciplines represented by the co-authors of that paper include mechanical engineering, computer science, electrical engineering, agronomy, and agricultural and biological engineering.

 

“Our work intersects with most of the colleges across campus,” said Asheesh Singh, the G.F. Sprague Chair in Agronomy at Iowa State and an associate dean for the College of Agriculture and Life Sciences. “We’re constantly seeding partnerships and collaborations.”

 

Those collaborations extend far beyond campus. They include other universities, businesses, industries, government agencies, farmers and even citizen-scientists who collect pest images. He said all these groups are creating data and launching “data co-ops” to share information.

 

“We have laid the foundation for this work, and it takes a village to build it,” said Ganapathysubramanian. “We’re very enthusiastic about all the skill sets required. This work has taken on a life of its own and it’s now an all-ISU effort.”

 

– 30 –

 

Read the papers

• “Machine Learning for High-Throughput Stress Phenotyping in Plants,” Trends in Plant Science, February 2016

• “Cyber-agricultural systems for crop breeding and sustainable production,” Trends in Plant Science, February 2024

• “InsectNet: Real-time identification of insects using an end-to-end machine learning pipeline,” PNAS Nexus, January 2025