Análise dos macronutrientes do solo sob enfoque multiespectral

• Ano de defesa: 2023
• Autor(a) principal: Paulo Roberto Nunes Viana
• Orientador(a): Rafael Felippe Ratke
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Análises. Algoritmos. Sustentabilidade. Solo.
• Link de acesso: https://repositorio.ufms.br/handle/123456789/5685

Analyzed soils lead to discussions in academic circles due to their possible contamination to the environment and lack of sustainability. To remedy these obstacles, the results obtained by traditional means of soil macronutrient analysis by chemometrics were studied and compared by soil analysis using a multispectral sensor (550 to 800 nm), in addition, the levels were classified of soil macronutrients by computational intelligence. In chapter 1, studies of predictive correspondence of results were performed using computational intelligence and decision algorithms Random Forest, M5P, Linear Regression, Multilayer Perceptron, Reptree and Random Tree. In chapter 2, the decision algorithms J48 and Logistic Regression were added, removing the M5P and Linear Regression, in addition to predicting the classification of quantitative grades compared in a soil analysis table in low, adequate, medium, high and its granulometry in clayey, sandy and medium and measure the harmony of the presented data verifying the precision of the experiments. It was possible to observe that for the analysis of sulfur a positive correlation coefficient of 66.82% and an absolute mean error of 8.22% were obtained, however for other macronutrients the prediction accuracy of the results obtained by the multispectral sensor compared to the chemometric analysis obtained predictive values little correlated . It was also observed that for the evaluation of phosphorus levels in the soil with the complementary addition of physical input variables of sand, silt and clay in the classifiers, it allowed an accuracy of 90% using the Random Forest classifier, while other classifiers presented relevant statistical differences showing import the technique used for both prediction and classification.