Vision and Olfactory-Based Wildfire Monitoring with Uncrewed Aircraft Systems

Wildfires threaten human lives, destroy facilities, and emit toxic smoke. Traditional wildfire monitoring methods are hindered by inflexibility (e.g., watch towers) and cannot provide precise geo-location of wildfires (e.g., satellites). Thanks to recent development in robotics, deploying uncrewed aircraft systems (UAS) to monitor wildfires has become a feasible solution. This article introduces a UAS-based wildfire monitoring system and implement it in a prescribed burn test. A multirotor UAS was employed as the search agent and carried both olfactory (i.e., carbon monoxide and particulate matter) and visual (i.e., a camera) sensors to detect the existence of wildfires. A fuzzy inference system is designed to fuse olfactory sensor outputs to estimate whether the UAS detects smoke. A deep learning model, i.e., You Only Look Once version 4 (YOLOv4), is employed to identify smoke from the captured images. We deployed the proposed UAS in a prescribed burn at Tallahassee, Florida, in May 2022. Experimental results show that the proposed fuzzy inference system improves the estimation accuracy of whether the UAS detects smoke compared with the fixed threshold algorithm. In addition, the proposed YOLOv4 model can also detect smoke from captured images with a small amount of training samples.