Caracterização do teor de nitrogênio foliar e produtividade do feijoeiro com técnicas de visão artificial

• Ano de defesa: 2008
• Autor(a) principal: Tormen, Gislaine Pacheco
• Orientador(a): Pinto, Francisco de Assis de Carvalho
• Banca de defesa: Fernandes Filho, Elpídio Inácio , Resende, Ricardo Capúcio de , Paula Júnior, Trazilbo José de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Engenharia Agrícola
• Departamento: Construções rurais e ambiência; Energia na agricultura; Mecanização agrícola; Processamento de produ
• País: BR
• Palavras-chave em Português: Agricultura de precisão; Visão artificial; Feijão
• Palavras-chave em Inglês: Precision agriculture; Machine vision; Beans
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA::MAQUINAS E IMPLEMENTOS AGRICOLAS
• Link de acesso: http://locus.ufv.br/handle/123456789/3524

Beans are one of the basic human nutrition components in Brazil and an important source of protein. Brazil is the major world producer and consumer, but has an average yield less than that of the USA and China. In the last years, the necessity to efficiently increase crop productivity and keep concerning with environmental issues has increased the producer interest in the use of new technologies such as precision agriculture techniques. The objective of this study was to evaluate the discrimination of leaf nitrogen (N) content classes using vegetation indices and chlorophyll meter measurements, and the discrimination of bean yield classes using the vegetation indices. The research considered two crop harvests ( dry crop and winter crop of 2007). The experiment was developed in randomized block design, with treatments in factorial scheme 5x6, with three replicates, summarize 90 plots. The treatments consisted of five different sowing N fertilization rates (0, 20, 30, 40 and 50 kg ha-1) and six different rates of topdressing N fertilization (0, 20, 30, 40, 60 and 80 kg ha-1) on urea composite. The vegetation indices were extracted from digital images, acquired using a system composed of two digital cameras. Therefore, the system acquired two images of the same scene at the same time (one in visible and the other in a near infrared spectral bands). The same leaves used to obtain the SPAD values were collected to determine the leaf N content. The leaf N content was sorted in low, medium and high classes. The yield was sorted in low, medium and high classes as well. In order to discriminate N and yield classes statistical classifiers were developed. To discriminate leaf N content classes, all possible combinations were used among the eight vegetation indices and SPAD values collected before topdressing fertilization. In order to discriminate yield classes, all possible combinations were used among the eight vegetation indices collected after topdressing fertilization. The chlorophyll SPAD measurements discriminated among the different rates of N applied on sowing in the two harvest seasons: in the first harvest season, at 25 day after emergence (DAE) and, in the second harvest, at 28 DAE. The SPAD value was correlated positively with leaf N content on the bean crop, having a greater correlation at 12 DAE. In the two crop harvests, the vegetation indices did not correlate with leaf N content values, but with the yield this correlation was positive and greater with the increase in days after emergence. In the first experiment, it was not possible to develop classifiers to discriminate leaf N content class, because the leaf N content values were higher than the considered tolerable levels, classifying all data into the high class. The use of the vegetation indices as characteristics vector was not useful on the leaf N content class discrimination, showing a low Kappa coefficient, classified as acceptable at 20 DAE and bad at 28 DAE, in the second experiment. When using SPAD measurements, the results improved, and Kappa coefficients were classified as good and very good at 20 and 28 DAE, respectively. Yield class discrimination obtained the greatest Kappa coefficient (44%) at 64 DAE in the first experiment, and, in the second experiment, the Kappa coefficient was greatest (76%) at 49 DAE. The vegetation indices were efficient in the discrimination of yield classes, and the combination of more than one vegetation index was important due to the variables group effect.