Assessing flood severity from crowdsourced social media photos with deep neural networks
- PDF / 3,045,546 Bytes
- 27 Pages / 439.642 x 666.49 pts Page_size
- 14 Downloads / 231 Views
Assessing flood severity from crowdsourced social media photos with deep neural networks Jorge Pereira1
˜ Monteiro2 · Joel Silva3 · Jacinto Estima3,4 · Bruno Martins1 · Joao
Received: 23 September 2019 / Revised: 20 May 2020 / Accepted: 8 June 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The use of social media data in disaster and crisis management is increasing rapidly. Particularly in connection to flooding events, geo-referenced images shared by citizens can provide situational awareness to emergency responders, as well as assistance to financial loss assessment, giving information that would otherwise be very hard to collect through conventional sensors or remote sensing. Moreover, recent advances in computer vision and deep learning can perhaps support the automated analysis of these data. In this paper, focusing on ground-level images taken by humans during flooding events, we evaluate the use of deep convolutional neural networks for (a) discriminating images showing direct evidence of a flood, and (b) estimating the severity of the flooding event. Considering distinct datasets (i.e., the European Flood 2013 dataset, and data from different editions of the MediaEval Multimedia Satellite Task), we specifically evaluated models based on the DenseNet and EfficientNet neural network architectures, concluding that these classification models can achieve a very high accuracy on this task, thus having a clear potential to complement other sources of information (e.g., satellite imagery) related to flooding events. Keywords Flood detection and severity estimation · Crowdsourced images · Image classification · Deep learning · Convolutional neural networks
1 Introduction The widespread use of mobile consumer electronics has made the act of taking and sharing photos online, for instance using smartphones or digital cameras coupled to GPS receivers, become commonplace. Information crowdsourcing through publicly shared photos has created a new opportunity to collect vast amounts of geo-referenced image data, which can for instance be useful in the context of disaster and crisis management. While crowdsourcing images posted on social media has been investigated in various contexts, leveraging such data in connection to flooding events (e.g., for detecting the extent Jorge Pereira
[email protected]
Extended author information available on the last page of the article.
Multimedia Tools and Applications
of the flooding or the water level in inundated areas) remains relatively unexplored, apart from a few recent initiatives [7, 45]. This new data collection methodology can have the advantage of providing a local perspective on inundations, whereas previous studies mostly rely on remotely sensed data from an overhead perspective [2, 4, 5, 11, 14, 27, 28, 34, 36, 38–40, 45, 47]. The local detail of crowdsourced images can perhaps provide useful information for estimating the boundaries of flood-water, including partial blockage of roadways due to flooding [7,
Data Loading...
