NIR Imagery-based Grass Fire Detection and Metrics Measurement using Small UAS

Abstract

This thesis focuses on the generation of a new grass fire aerial image dataset and development of novel methods for near-infrared (NIR) imagery-based fire front identification and fire depth estimation using small unmanned aircraft systems (sUAS). The procedure for collection and creation of the Grass Fire Front and near-Infrared (NIR) and Thermal Imagery (GRAFFITI) dataset is introduced first including two levels of data: synced raw thermal and red, green and near-infrared (RGNIR) image pairs and processed image pairs of the same overlapping field-of-view. A novel NIR imagery-based fire detection and fire front identification algorithm is then proposed and validated against manually labeled ground truth, using the GRAFFITI dataset. A comparative study is further performed on the problem of grass fire front location and flame depth estimation using thermal and NIR imagery. Finally, recommendations are made to future researchers who are interested in wildland fire sensing using thermal or NIR imagery.