When working on his latest science fair project, Robin Yadav says he wanted to use his knowledge of math and computer science “to see if I can help the wildfires in B.C.”
Robin, a Grade 12 student at Queen Elizabeth Secondary, created a project, “Deep Learning based Fire Recognition for Wildfire Drone Automation,” that uses machine learning and object detection to perform video-based fire recognition using a drone.
His project won the Intact Financial Climate Change Resilience Award at the nationwide Youth Science Canada online STEM Fair, as well as the BC Game Developers Innovation and Engineers and Geoscientists awards for the South Fraser region at the BC/Yukon Virtual Science Fair.
Robin said he always been “really interested” in math, and by the time he was in Grade 9, he’d completed his high school math requirements.
“I just wanted to see what else I could do with the math knowledge I have. So over the summer, I started learning machine learning and computer science and coding and all that. At that time, I was thinking I wanted to do a science fair project,” he said.
He said he wanted to see if he could use his knowledge of math and computer science to potentially help with fighting wildfires in B.C.
“My original idea revolved around could be to possibly harness the energy that’s dissipated by these wildfires,” he explained. “But then I started researching more into wildfires and I found that drones are becoming pretty useful tools for firefighters to use when they’re attacking wildfires.
“I decided, let’s automate drones using machine learning and object detection, which is recognizing fire in the video footage of drones.”
From there, Robin said he contacted fire departments in B.C. and Silicon Valley “because their drone usage is more advanced just because of the tech over there.”
“In B.C., they use it more for monitoring purposes after the fire is mostly under control, to locate hot spots, to check the fire front and see if the fire is being properly contained,” he said.
“Projects like these or ideas like these really need continual development, so it’s not just one thing and you’re finished. They can always be upgraded, new features can be added, so it’s going to be more of a longer development process.”
Robin said automation “is a pretty difficult process, and I realized that traditional… techniques won’t necessarily work, so I decided to go into the artificial intelligence and machine learning route.”
“What you essentially do is train machine learning models to recognize objects in its field of view, so I trained the drone — or specifically the program that the drone runs on — to perceive the fire in its video feed,” he said.
“I collected around 6,000 images of fire and then labelled all of them, ‘Here’s a fire in this image, here’s a fire in this image’ and then I trained the model and after a while, the model performs better and better and better.”
Going forward, Robin said he plans to do some field testing this year before taking the project to fire departments in the province, adding there are still some improvements to be done.
“There’s extensive knowledge in firefighting in British Columbia, so… I’m seeing if I could get local fire departments could do some testing with the project I created.”
Like us on Facebook Follow us on Instagram and follow Lauren on Twitter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .