AMES, Iowa – It isn’t always easy to know what insect you’re looking at. An app called InsectNet is hoping to eliminate those challenges, providing farmers with an easy way to identify pests in the field.

InsectNet is simple to use. Like other apps that identify plants, animals, or insects, users are asked to upload a photo of what they’re observing, and the app will provide an identification or prediction for what the species may be. The global model has insects of all backgrounds: pests, predators, pollinators, indicator species and more.

InsectNet is backed by a dataset of 12 million insect images and can identify and predict more than 2,500 insect species. The app is a product of machine learning and artificial intelligence; with training, the app “learns” what certain species look like so that when it’s presented with an image, it can predict what the insect is.

The current model runs at 96 percent accuracy; when the model is unsure of the species, it provides users with the likelihood of several identifications. For example, it might say it’s 80 percent sure the insect in the photo is a western conifer seed bug, but it’s 20 percent sure the insect is a brown marmorated stinkbug. The app can also be fine-tuned to a particular local region.

The team collected around 30,000 images of different kinds of insects that impact corn, soy, alfalfa and vegetable crops. The app also sources some of its data and images from iNaturalist, a popular application used to identify flora and fauna worldwide.

What makes InsectNet unique is the focus on agriculture and farmers.

“There are tools that can identify insects, but those are mostly for your garden or when you’re out for a hike – they’re not specifically agronomically relevant,” Baskar Ganapathysubramanian, professor of mechanical engineering at Iowa State University, said. “(InsectNet) is a robust tool that farmers, breeders, growers, and general enthusiasts can use to rapidly identify insects and figure out if it’s a pest or a predator or whatever its role is in the agricultural system.”

Ganapathysubramanian also added that when gardeners and hobbyists use apps like iNaturalist, making a misidentification isn’t a big deal. For farmers, misidentifying an insect can be a costly mistake, especially in regions like the Midwest that only have one growing season.

Arti Singh, associate professor in the department of agronomy at ISU, said that she was originally interested in using AI and machine learning to predict plant diseases. She wanted to know if machine learning could identify, classify, quantify, or predict plant stress ahead of time. Unfortunately, there was insufficient data to train the model at the time.

“But then in 2020, I was working on a lot of new crops, and my crops were hit hard by insects,” Singh said. “I kept thinking (the crop was suffering from) high heat stress or flower drop, but by the time I came to know (it was insects), it was too late, and the crop was lost. I said to Baskar, ‘I think we have to work on insects at this point.’”

Singh believes that InsectNet could be an important tool in pest management strategies, especially with the addition of a chatbot. A chatbot component would provide farmers with information about the insect they identified and what steps can be taken to manage it. Additionally, the team is working on making the model more robust so that it can identify insects in various stages of life, including egg and larval stages.

Identifying insects at an egg stage can be useful for curbing the spread of invasive species like the spotted lanternfly, which hasn’t established in Iowa. InsectNet could also be useful for officials working at ports of entry where invasive species often cross borders undetected. Authorities could potentially use InsectNet for any suspicious insects on products.

Ganapathysubramanian said that the team is working on making changes to the model’s architecture to improve accuracy. Additionally, researchers want to make the app available on phones for easier accessibility and potentially add the technology to drones.

“What we would like to do is deploy (InsectNet) on robotic platforms or drones that can fly through fields and create hotspots where it’s seeing insects,” he said. “So, it’s not about whether you find an insect or not, but about an action threshold. This is potentially transformative for growers because if you are able to identify hotspots, a farmer can choose to only spray in those hotspots instead of spraying the entire field.”

The app is currently publicly available at https://insectapp.las.iastate.edu/