Utilizing Post-Hurricane Satellite Imagery to Identify Flooding Damage with Convolutional Neural Networks

TL;DR

This paper demonstrates that convolutional neural networks can accurately classify post-hurricane satellite images to identify flooded or damaged buildings, offering a faster alternative to traditional field assessments.

Contribution

It introduces a CNN-based approach for post-hurricane damage detection using satellite imagery, achieving over 99% accuracy, which advances remote sensing disaster assessment methods.

Findings

CNN models achieved over 99% accuracy in damage classification.

Deep learning provides a rapid, reliable alternative to field reconnaissance.

The approach is validated on post-Hurricane Harvey satellite data.

Abstract

Post-hurricane damage assessment is crucial towards managing resource allocations and executing an effective response. Traditionally, this evaluation is performed through field reconnaissance, which is slow, hazardous, and arduous. Instead, in this paper we furthered the idea of implementing deep learning through convolutional neural networks in order to classify post-hurricane satellite imagery of buildings as Flooded/Damaged or Undamaged. The experimentation was conducted employing a dataset containing post-hurricane satellite imagery from the Greater Houston area after Hurricane Harvey in 2017. This paper implemented three convolutional neural network model architectures paired with additional model considerations in order to achieve high accuracies (over 99%), reinforcing the effective use of machine learning in post-hurricane disaster assessment.