Web-based apps identify insects around the world and around the farm

Farmers notice insects that are not familiar with leaves. Is this a pollinator? Or pests? Good news when harvesting? Or bad? Do you need to control it? or not?

The farmer can take photos, use his smartphone or computer to supply photos to a web-based application called Insectnet, and regain real-time information with the help of machine learning technology.

“The app identifies insects and returns predictions of taxonomic classification and role in ecosystems as pests, predators, pollinators, parasites, degradators, herbivores, indicators, and invasive species.” pnas nexus. Bascargana Patish Bramanian and Artisin of Iowa State University are their corresponding authors.

INSECTNET is supported by a dataset of 12 million insect images, including many collected by citizen scientists, but provides identification and prediction of over 2,500 insect species with an accuracy of over 96%. If the application is not sure about the insect, it says it is uncertain and gives users more confidence when providing answers.

Additionally, applications were built as local models from global, allowing geographical tweaks using local and regional datasets tested by experts. It is useful for farmers everywhere.

Therefore, beware of Army insects, cutworms, grasshoppers, stink bugs, and all other harmful insects. And hello, butterflies, bees and all the other pollinators. I’m happy to meet you, the Lady Beetles, the Manz and all the other pest predators.

“We envision that we will complement existing approaches and become part of a growing suite of AI technologies to address agricultural challenges,” the author writes.

Researcher’s Village

Singh, an associate professor of agriculture, said the ability of INSECTNET to fine-tune to a particular region or country is particularly useful.

In Iowa, for example, Singh said there are around 50 insect species that are particularly important for the state’s agricultural production. To identify and provide predictions about these insects, Singh said that around 500,000 insect images were used in the project.

It can happen to farmers all over the world. Additionally, if there is not enough data, these sophisticated models may require millions of images, but for local tweaks, the global dataset is available to farmers.

However, Insectnet is not just farmers. Singh said it will also help port agents and border crossing agents identify invasive species. Or it could be useful for researchers working on ecological research.

Therefore, the app is usable and flexible. But is it accessible?

Joseph and Elizabeth Anderrick, professor of engineering at the Iowa-based AI Resilient Agriculture for Resilient Agriculture, Ganapathysubramanian, is still going to the App Store to download the version. cannot. However, the app runs on servers in Iowa. Using a QR code (see sidebar) or this URL, users can upload insect photos to get identification and predictions.

This works throughout the stages of insect life, from eggs to larvae to larvae to adults. Works with the seeds that look like they are. And it works with a variety of image quality and orientation.

The user’s last line is basic information about insects. “Is this a pest?” Shin said. “Or are you friends?”

The developer demonstrated the app at the Farm Progress Show last August in Boone, Iowa. And now, research papers introduce it to a broader, scientific audience.

But are there already apps to help identify insects?

Yes, Ganapathysubramanian said, but they are not the scale of insect nets and are not capable of global to local applications. It is also not an open source application with technology to share.

“Making open source for INSECTNET can encourage broader scientific efforts,” he said. “The scientific community can build on these efforts, rather than starting from scratch.”

The project also answered many technical questions that could be applied to other projects, he said.

How much data is sufficient? Where can I get that much of the data? What can you do with noisy data? How much computer power do you need? How do you deal with a lot of data?

“Finally, you need a village of expertise to get to this point, right?” Ganapatishbramanian said.

Agriculturalists and computer engineers, statisticians, data scientists, artificial intelligence experts, and artificial intelligence experts have made INSECTNET work together for about two years.

“What we’ve learned can include expanding insects to include weed and plant diseases, or other related identification and classification issues in agriculture,” Singh said. “It’s very close to a one-stop shop to identify all of this.”