Automatic detection of potential mosquito breeding sites from aerial images acquired by unmanned aerial vehicles. (November 2021)

• Record Type: Journal Article
• Authors: Bravo, Daniel Trevisan; Lima, Gustavo Araujo; Alves, Wonder Alexandre Luz; Colombo, Vitor Pessoa; Djogbénou, Luc; Pamboukian, Sergio Vicente Denser; Quaresma, Cristiano Capellani; Araujo, Sidnei Alves de
• Is Part Of: Computers, environment and urban systems. Volume 90(2021)
• Publication Date: 2021-11
• Subjects: Vector control -- Mosquito -- Unmanned aerial vehicle -- Objects detection -- Convolutional neural network -- Support vector machine -- Bag of visual words; City planning -- Data processing -- Periodicals; Regional planning -- Data processing -- Periodicals; 303.4834
• DOI: 10.1016/j.compenvurbsys.2021.101692
• Languages: English
• ISSNs: 0198-9715

Abstract

Abstract: The World Health Organization (WHO) has stated that effective vector control measures are critical to achieving and sustaining reduction of vector-borne infectious disease incidence. Unmanned aerial vehicles (UAVs), popularly known as drones, can be an important technological tool for health surveillance teams to locate and eliminate mosquito breeding sites in areas where vector-borne diseases such as dengue, zika, chikungunya or malaria are endemic, since they allow the acquisition of aerial images with high spatial and temporal resolution. Currently, though, such images are often analyzed through manual processes that are excessively time-consuming when implementing vector control interventions. In this work we propose computational approaches for the automatic identification of objects and scenarios suspected of being potential mosquito breeding sites from aerial images acquired by drones. These approaches were developed using convolutional neural networks (CNN) and Bag of Visual Words combined with the Support Vector Machine classifier (BoVW + SVM), and their performances were evaluated in terms of mean Average Precision - mAP-50. In the detection of objects using a CNN YOLOv3 model the rate of 0.9651 was obtained for the mAP-50. In the detection of scenarios, in which the performances of BoVW+SVM and a CNN YOLOv3 were compared, the respective rates of 0.6453 and 0.9028 were obtained. These findings indicate that the proposed CNN-based approaches can be used to identify potential mosquito breeding sites from images acquired by UAVs, providing substantial improvements in vector control programs aiming the reduction of mosquito-breeding sources in the environment. Highlights: Computational approaches for the automatic identification of potential mosquito breeding sites were proposed The approaches were developed using state-of-the-art techniques (CNN, Bag of Visual Words and SVM classifier) The proposed CNN-based approaches obtained good results in the detection of suspected objects and scenarios The proposed UAV-based methods allow for constant monitoring, providing data with high spatial and temporal resolutions The proposed UAV-based methods can considerably help to increase the community involvement in vector control programs