Detecção de podridão mole em alface por Pectobacterium carotovorum subsp. carotovorum por algoritmos de aprendizado de máquina a partir de imagens multiespectrais

• Ano de defesa: 2021
• Autor(a) principal: Carmo, Glecia Júnia dos Santos
• 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: Classificação supervisionada; Supervised classification; Diagnose de doença; Diagnosis disease; Lactuca sativa; Veículo aéreo não tripulado; Unmanned aerial vehicle; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::GEOFISICA::SENSORIAMENTO REMOTO; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::FITOSSANIDADE::FITOPATOLOGIA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/32619; http://doi.org/10.14393/ufu.di.2021.354

Lettuce is the most consumed leafy vegetable and has the greatest economic importance worldwide, however production is hampered by numerous pathogens, including Pectobacterium carotovorum subsp. carotovorum, which causes soft rot in lettuce. Thus, this study aimed to identify the best sensor and determine the optimal stage to detect disease caused by Pectobacterium carotovorum subsp. carotovorum in lettuce, using images obtained by multispectral sensors mounted on an unmanned aerial vehicle (UAV). For this, an experiment was installed in a greenhouse at the Federal University of Uberlândia, Campus Monte Carmelo, containing 392 plastic pots of five liters each, and in each pot a lettuce seedling was transplanted. At 27 days after transplanting, 196 plants were inoculated with Pectobacterium carotovorum subsp. carotovorum and 196 plants were not inoculated. At 4, 8, 12, 16, 20, 24 and 28 days after inoculation (DAI), lesions of plants inoculated were assessed and plants inoculated or not were assessed for agronomics parameters. At the same time interval, flights were also carried out in the area, between 12 and 13 hours. Were used the Support Vector Machine (SVM) and Naive Bayes (NB) classifiers to analyze data groups consisting of spectral bands, vegetation indices and a combination of bands and indices obtained from a conventional visible camera and Mapir Survey3W multispectral camera, as well as agronomic parameters. The results confirmed the possibility of pre-symptomatic detection of Pectobacterium carotovorum subsp. carotovorum in lettuce. With respect to identifying infected lettuce plants by supervised classification, the best results were obtained at 4 and 8 DAI, especially when using the subsets derived from the Mapir Survey3W camera (RGN sensor), for both classifiers. The subsets obtained with the conventional visible sensor (RGB sensor) produced the best results at 20 and 24 DAI.