[Paper Review] Disaster Monitoring using Unmanned Aerial Vehicles and Deep Learning

Andreas Kamilaris, Francesc X. Prenafeta‐Boldú|arXiv (Cornell University)|Jul 31, 2018

Video Surveillance and Tracking Methods28 references44 citations

TL;DR

The paper demonstrates that a simple deep learning model applied to UAV-captured aerial images can identify disasters with 91% accuracy on a 544-image dataset.

ABSTRACT

Monitoring of disasters is crucial for mitigating their effects on the environment and human population, and can be facilitated by the use of unmanned aerial vehicles (UAV), equipped with camera sensors that produce aerial photos of the areas of interest. A modern technique for recognition of events based on aerial photos is deep learning. In this paper, we present the state of the art work related to the use of deep learning techniques for disaster identification. We demonstrate the potential of this technique in identifying disasters with high accuracy, by means of a relatively simple deep learning model. Based on a dataset of 544 images (containing disaster images such as fires, earthquakes, collapsed buildings, tsunami and flooding, as well as non-disaster scenes), our results show an accuracy of 91% achieved, indicating that deep learning, combined with UAV equipped with camera sensors, have the potential to predict disasters with high accuracy.

Motivation & Objective

• Motivate the use of UAVs with camera sensors for disaster monitoring and mitigation.
• Evaluate the potential of deep learning to recognize disaster events from aerial imagery.
• Highlight a relatively simple DL model that achieves high accuracy on disaster identification.
• Demonstrate the applicability of this approach to multiple disaster types (fires, earthquakes, collapsed buildings, tsunamis, floods).

Proposed method

• Use unmanned aerial vehicles equipped with cameras to collect aerial imagery of disaster and non-disaster scenes.
• Apply a deep learning model to the dataset to perform disaster identification from images.
• Evaluate recognition performance and report accuracy on the dataset.
• Discuss the potential of combining UAV sensing with deep learning for rapid disaster prediction and response.

Experimental results

Research questions

• RQ1 Can deep learning on UAV-acquired aerial images accurately identify disaster events?
• RQ2 What is the achievable accuracy of a DL model on a 544-image dataset of disasters and non-disasters?
• RQ3 How well does this approach generalize across different disaster types (fires, earthquakes, collapsed buildings, tsunami, flooding)?

Key findings

• On a dataset of 544 images including disasters and non-disaster scenes, the model achieved 91% accuracy.
• The study demonstrates the feasibility of a relatively simple deep learning model for disaster identification using UAV imagery.
• The results indicate high potential for UAV-based DL systems to aid disaster monitoring and rapid response.