Detecção de Planococcus citri em cafeeiro por imagens multiespectrais

• Ano de defesa: 2022
• Autor(a) principal: Rossati, Kamila Fernanda
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Agricultura e Informações Geoespaciais
• Palavras-chave em Português: aprendizado de máquina; cafeicultura; cochonilha-da-roseta; detecção de pragas; imgens multiespectrais; machine learning; coffee growing; rosette mealybug; pest detection; multiespectral imaging; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::FITOSSANIDADE::ENTOMOLOGIA AGRICOLA; Agronomia; Café - Cultivo; Cochonilha - Imagens multiespectrais; Café - Doenças e pragas
• Link de acesso: https://repositorio.ufu.br/handle/123456789/36077; http://doi.org/10.14393/ufu.di.2022.425

Coffee growing provides around 125 million jobs. In the coffee plant, the mealybug Planoccous citri is found in reboleiras, and can infest branches, leaves, flower buds and fruits, reaching the point of infesting all the rosettes of the plant and leading to partial stunting or total loss of these branches. Sampling is not performed due to the cost and time required for this practice, so the applications occur in the total area. In this way, it is necessary to introduce technologies that allow the provision of information for better decision-making in its management. The objective of this study was to verify the potential of using low-cost multispectral images in the discrimination of coffee plants infested by P. citri. Three study areas were used, the first in the municipality of Coromandel, MG, one with a high infestation of cochineal and the second with no cochineal and an area in the municipality of Monte Carmelo, MG, with presence and absence of cochineal in the evaluated plants. . In each study area, 50 plants were randomly sampled, with a minimum distance of 10 meters between plants, evaluating the amount of mealybugs present in 2 plagiotropic branches located in the middle third of the plants. The images were obtained using a drone coupled to a Mapir Survey 3W camera at a height of 100 meters. The classifications were made using the algorithms Artificial Neural Networks (ANN), Support Vector Machine (SMO) and Random Forests. The results confirmed the possibility of discrimination between healthy and P. citri infested plants using algorithms based on machine learning. Regarding the discrimination of healthy and infested plants, the Random Forest algorithm showed the best result in areas with infestation variability (EG=90% and K=0.80), followed by SMO (EG=83.34% and K= 0.67) and RNA (EG=73.34% and K=0.47).