Fighting Wildfires with Computer Vision

Abhishek Ghosh, a graduate student at Texas A&M University, is developing a computer vision model to detect early signs of forest fires by distinguishing smoke from clouds, with the aim of deploying drones to extinguish fires before they escalate. As part of a collaboration with AI For Mankind, the project has open-sourced a wildfire smoke detection dataset under a BY-NC-SA 4.0 license to assist other researchers. Ghosh chose the EfficientDet D0 model after experimenting with various options and used Roboflow Pro for advanced augmentations to enhance model performance, overcoming challenges like differentiating smoke from clouds by adjusting annotations and cropping images. While the model currently processes historical data, the goal is to implement it in real-time using drones to cover broader areas at risk of fires, with the vision of deploying a drone fleet to proactively detect and extinguish nascent fires.