Water tank and swimming pool detection based on remote sensing and deep learning: Relationship with socioeconomic level and applications in dengue control

Studies have shown that areas with lower socioeconomic standings are often more vulnerable to dengue and similar deadly diseases that can be spread through mosquitoes. This study aims to detect water tanks installed on rooftops and swimming pools in digital images to identify and classify areas base...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:PloS one 2021-01, Vol.16 (12), p.e0258681
Hauptverfasser: Higor Souza Cunha, Brenda Santana Sclauser, Pedro Fonseca Wildemberg, Eduardo Augusto Militão Fernandes, Jefersson Alex Dos Santos, Mariana de Oliveira Lage, Camila Lorenz, Gerson Laurindo Barbosa, José Alberto Quintanilha, Francisco Chiaravalloti-Neto
Format: Artikel
Sprache:eng
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Studies have shown that areas with lower socioeconomic standings are often more vulnerable to dengue and similar deadly diseases that can be spread through mosquitoes. This study aims to detect water tanks installed on rooftops and swimming pools in digital images to identify and classify areas based on the socioeconomic index, in order to assist public health programs in the control of diseases linked to the Aedes aegypti mosquito. This study covers four regions of Campinas, São Paulo, characterized by different socioeconomic contexts. With mosaics of images obtained by a 12.1 MP Canon PowerShot S100 (5.2 mm focal length) carried by unmanned aerial vehicles, we developed deep learning algorithms in the scope of computer vision for the detection of water tanks and swimming pools. An object detection model, which was initially created for areas of Belo Horizonte, Minas Gerais, was enhanced using the transfer learning technique, and allowed us to detect objects in Campinas with fewer samples and more efficiency. With the detection of objects in digital images, the proportions of objects per square kilometer for each region studied were estimated by adopting a Chi-square distribution model. Thus, we found that regions with low socioeconomic status had more exposed water tanks, while regions with high socioeconomic levels had more exposed pools. Using deep learning approaches, we created a useful tool for Ae. aegypti control programs to utilize and direct disease prevention efforts. Therefore, we concluded that it is possible to detect objects directly related to the socioeconomic level of a given region from digital images, which encourages the practicality of this approach for studies aimed towards public health.
ISSN:1932-6203
DOI:10.1371/journal.pone.0258681