Wildfire Detection Using CNNs Analyzing the effect of Multispectral and Hyperspectral Imaging from Landsat-8

Keywords: Wildfire, wildfire detection, satellite, satellite images, Spectral bands, infrared, near-infrared, CNN, U-Net, SWIR, TIR
Abstract Type: Poster Abstract

Abstract

Wildfires peak globally during the months of August and September, posing a substantial threat to ecological and social environments. To address these challenges, remote sensing technologies have gained prominence amid the global impact of wildfire occurrences. Our study will investigate the efficacy of Convolutional Neural Networks (CNNs) in analyzing multispectral and hyperspectral imaging for detecting wildfire signatures.

Our approach extends previous research by emphasizing the selection and preprocessing of spectral bands sensitive to thermal anomalies and vegetation changes for isolating fire signatures. To then create a comprehensive dataset encompassing various spectral bands, critical for an in-depth analysis of their contributions to wildfire detection. We employ a U-Net CNN architecture to process these spectral band images and focus specifically on the bands sensitive to thermal anomalies and vegetation changes, such as the thermal infrared (TIR) and near-infrared (NIR) bands. Each model will be trained on patches of processed images and masks generated by well-established fire detection algorithms.

The U-Nets will be configured with consistent architectural parameters such as convolutional filters and input channels to assess their adaptability across diverse spectral band combinations. Performance will be evaluated using precision, recall, and F1-score to identify the most effective wildfire detection approach. We anticipate that thermal infrared and shortwave infrared (SWIR) bands will substantially improve fire detection, while visible light bands will have a lesser impact. This study underscores the importance of selecting appropriate spectral bands in the preprocessing stage to maximize the effectiveness of CNNs in wildfire monitoring and response.