Sistema de visão computacional para identificação automática de potenciais focos do mosquito Aedes aegypti a partir de imagens adquiridas por drones

• Ano de defesa: 2021
• Autor(a) principal: Lima, Gustavo Araujo
• Orientador(a): Araújo, Sidnei Alves de
• Banca de defesa: Araújo, Sidnei Alves de , Quaresma, Cristiano Capellani , Sassi, Renato José , Belan, Peterson Adriano
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Nove de Julho
• Programa de Pós-Graduação: Programa de Pós-Graduação em Informática e Gestão do Conhecimento
• Departamento: Informática
• País: Brasil
• Palavras-chave em Português: drone; Aedes aegypti; reconhecimento de padrões; visão computacional; redes neurais convolucionais; YOLO
• Palavras-chave em Inglês: drone; Aedes aegypti; pattern recognition; computer vision; convolutional neural networks; YOLO
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://bibliotecatede.uninove.br/handle/tede/3045

The World Health Organization (WHO) has warned that effective vector control measures are essential to reduce the incidence of infectious diseases transmitted by mosquitoes, such as Aedes aegypti. In this sense, unmanned aerial vehicles (UAVs), popularly known as drones, have become an important technological tool for health surveillance teams to map and eliminate mosquito breeding sites in areas where diseases such as dengue, Zika, chikungunya and malaria are endemic, since they allow the acquisition of aerial images with high spatial and temporal resolution. However, these images are usually analyzed by manual processes that consume a lot of time in the interventions of control and combat mosquitoes. This work presents an investigation about the applicability, limitations and scalability of approaches found in the literature for automatic identification of potential mosquito breeding sites using drones, as well as the development of a computer vision system (CVS) for this purpose, which is able to indicate through georeferenced data, the location of suspicious objects and scenarios in the aerial images acquired by a drone. To this end, different convolutional neural networks (CNN) configurations from the YOLO v4 framework were implemented and evaluated, which presented hit rates and mAP-50 (mean average precision) ranging from 0,8926 and 0,9061 to 0,9513 and 0,9629. To conduct the CNN’s evaluations, we composed a database of 500 images acquired in urban areas of the Metropolitan Region of São Paulo (MRSP), including objects and scenarios defined as targets. The results obtained, compared to recent results in the literature, indicate that the CNN-based approach was adequate to compose the proposed CVS which gave rise to a software artifact, and that the use of drones can provide substantial improvements in programs to prevent and combat mosquito breeding sites.