Georgia Tech Researchers Monitor Campus Biodiversity with Machine Learning |

Georgia Tech Researchers Monitor Campus Biodiversity with Machine Learning

February 28, 2022

By Selena Langner

Since June, Lalith Polepeddi and Akhil Chavan have been using their skills in computer science and machine learning to help study biodiversity in Georgia Tech’s new EcoCommons.

Both research staff at the Georgia Tech Global Change Program, Polepeddi and Chavan teamed up to apply for a micro research grant from the Kendeda Living Building last summer. The grants empower research and innovation at a student, staff, and faculty level through small, accessible, amounts of seed funding.

Their goal? To automatically identify animal species visiting the EcoCommons using machine learning.

“We were excited about this project because building a regenerative environment on campus goes beyond sustainability,” says Chavan. “We can utilize machine learning models to better understand which species are living on our campus, and then share that information so we can collaborate with others to strengthen the habitats of native species here.

Georgia Tech’s new EcoCommons are a haven for biodiversity. Here, a trap camera catches a bird in song. Click “play” to watch the 10-second video.

Over the last six months, Polepeddi and Chavan built a pipeline that uses pre-trained machine learning models to classify animal species in video footage from camera traps placed around Georgia Tech’s campus.

“The pipeline takes as input a dataset of video recordings gathered from different locations on GT campus, and splits each video into ~10 separate images,” Polepeddi explains. “From there, a pre-trained model runs over each image to detect whether an animal is visible. If an animal is present, another pre-trained model runs over the image to classify the species of the animal.”

Animals are identified by the system in two key steps: first, the AI model identifies if an animal is present in the footage. Then, another model classifies the animal. The system allows researchers to become more familiar with biodiversity on campus. “There are a few feral cats, and potentially a fox, too, which hasn’t been caught on camera yet,” Chavan says.

The project builds on, and is in collaboration with, ongoing work by Dr. Emily Weigel, a Senior Academic Professional in the School of Biological Sciences who is monitoring biodiversity on campus. Polepeddi and Chavan hope their work creating an automated system to classify animals can save Weigel’s team time and resources that might otherwise be spent searching through hours of camera footage to identify individual animals manually.

In turn, the work by Weigel’s team placing cameras and microphones in the EcoCommons and gathering a dataset of images and audio recordings of species proved instrumental in streamlining Polepeddi and Chavan’s.

“We were expecting that setting up cameras and capturing footage of animals would be a significant task for us.” says Chavan. “But by teaming up with Dr. Emily Weigel and her research group, we were able to jump right into building a program to classify the animal species in video footage that Dr. Weigel’s team had already gathered using camera traps around campus.”

Polepeddi and Chavan were one of five teams awarded a Kendeda Building Micro-Grant. Proposals ranged from quantifying per-capita single-use waste impact, to developing a deep-learning method for recognizing invasive plant species, to creating a campus pollinator garden. One key goal of the micro-grant initiative is to seed cross-campus collaboration and innovation in using Georgia Tech’s campus as a living laboratory.

Trap cameras provide a wealth of data, including temperature, time, and date. Coupled with a positive animal ID, they can be used to better understand patterns in Georgia Tech’s biodiversity.

The Kendeda Building Micro-Grants emphasize that anyone on Georgia Tech’s campus, whether staff, student, or faculty, can help work toward a more sustainable future. One of the team’s key takeaways? “Although we didn’t have much experience with studying biodiversity, by applying machine learning techniques to this project, we were able to utilize our skills to explore a new field while adding value to an ongoing project.”

All awardees will present their findings at the first micro-grants research conference, which will be held during Georgia Tech’s first-annual SDG Week in March 2022.