����ֱ��

The Buzz on AI

Computing Sciences professor collaborates with entomologists to research pollinators

By Yasmine Iqbal

Venkat Margapuri, PhD, assistant professor in the Department of Computing Sciences, uses computer vision and artificial intelligence to better understand the natural world. He develops convolutional neural networks (CNNs), a type of computer architecture modeled on the human visual system that can learn features and patterns from data and images. At Kansas State University, where he received his doctoral degree, Dr. Margapuri used CNNs to study a host of things, from seeds to galaxies.

He’s now collaborating with colleagues at Kansas State to collect and analyze images of Midwestern prairies. The researchers’ goal is to study the habitat and prevalence of a group of essential insects: bumble bees.

Native bumble bees (genus Bombus) are some of the most effective pollinators of wildflowers, as well as many agricultural crops, including tomatoes, melons and blueberries. There are roughly 50 species of bumble bee in the US, but their numbers are declining due to multiple factors, including habitat destruction and pesticide use.

Accurately identifying bumble bees in their native habitats used to require capturing the insects to closely examine them. This taxonomic work is now image-based, digitized and non-invasive. Researchers and the general public can contribute to and access massive image repositories, including the Global Biodiversity Information Facility (GBIF).

Dr. Margapuri’s project, which is supported by a grant from the US Department of Agriculture’s National Institute of Food and Agriculture, will involve flying autonomous aerial vehicles over Kansas prairies to take photos and videos of bumble bees and the nearly 200 flowering plants they forage upon. Dr. Margapuri and his team will then use the images to train the CNN to recognize and classify different species.

“The images in digital repositories like GBIF are mostly close-up, side-view shots,” Dr. Margapuri explains. “We’ll be using the aerial images to ‘fine-tune’ the CNN’s ability to recognize these organisms from all angles, with varying resolutions and lighting conditions.”

Garik Kazanjian ’27 CLAS, a student in Dr. Margapuri’s lab, is helping to train the CNN to make it faster and more efficient. “We’ll be able to leverage this technology to better understand which plants are correlated with which bumble bee species,” he says.

This work will also enhance the BeeMachine app, developed by Kansas State researchers, which uses computer vision to recognize more than 350 types of insects. “Expanding BeeMachine to help identify the bees’ food sources will help us understand how much forage is needed to sustain healthy pollinator populations,” Dr. Margapuri says. “It will also make it a more robust tool for scientists and the general public.”