Satellite Imagery-Based Wildfire Detection Using Deep Learning

Abstract

Satellite imagery is not currently used in real-time, or even near real-time, to add to forest fire prevention. In this chapter, we travel around the use of one significant worldwide satellite – Sentinel-2 – to check the detection of wildfire. All earlier works in literature are focused on the use of Landsat (30–60 m resolution), whose satellite imagery service is less granular. To detect these wildfires, we are extracting different multiple features like vegetation index, true color, and moisture content via the Sentinel satellite (10–40 m). Each day, these active fire product outputs a data set that holds hundreds of “detected fires” at given latitude/longitude coordinates. Using deep learning, we can tease out which of these “detected fires” are forest fires and use this knowledge in near real-time to aid in forest fire prevention. Hundreds of forest fires burned over 9 million acres of land in 2015, causing millions of dollars in property damage and immeasurable loss and pain to those families affected. Understanding, tracking, and effectively fighting forest fires are crucial in terms of minimizing this damage and loss. Using satellite imagery of potentially detected forest fires can greatly aid in this process. Since the amplified incidences and harshness of wildfires AU: Please check edits in this sentence. in present times, for a better wildfire detection process easing, the minimization of response delay has become vital. In order to address these challenges, modern progress in the field of deep learning and computer vision unitedly can be leveraged with the rise of frequent satellite representation. Hereby we propose AU: Please check edits in this sentence. a system of deep learning wildfire detection model that uses locations of excessive fire images and is trained from the Sentinel-2 satellite.