Fenotipagem não destrutiva usando espectroscopia no infravermelho próximo e quimiometria em sementes de mamona

• Autor(a) principal: Santos, Maria Betania Hermenegildo dos
• Data de Publicação: 2015
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFPB
• Texto Completo: https://repositorio.ufpb.br/jspui/handle/tede/7164
• topic: Semente; Mamoneira; Ricina; Espectroscopia NIR; Calibração multivariada; Seed; Castor bean; Ricin; NIR spectroscopy; Multivariate calibration; CIENCIAS EXATAS E DA TERRA::QUIMICA

In this work we used the near infrared spectroscopy (NIR) and chemometric tools to develop e classification models of two different cultivars of castor bean BRS Nordestina (N) and BRS Paraguaçu (P). It was also studied the feasibility of calibration models for ricin content in seeds prediction of three cultivars of castor bean (BRS Nordestina, BRS Paraguaçu and BRS Energia). Diffuse reflectance spectra were recorded in the region of 400-2500 nm. For classification models were used 350 intact seeds for each cultivar. In the calibration sample set was formed by 69 scarified seeds, 25 of BRS Energia, 25 of BRS Nordestina and 19 of BRS Paraguaçu. Measurements were made at four positions for each seed. The spectra are pre-processed with Savitzky-Golay algorithm with a 15 points window, first derived for baseline correction. Based on PCA (Principal Component Analysis) models, the region corresponding to the spectral range from 2110 to 2155 nm, was selected because it has good distinction between cultivars. SIMCA (Soft Independent Modeling of Class Analogy) model provided promising results in the classification of seed for the significance levels 1, 5 and 10%. The SPA-LDA (Sucessive Projections Algorithm-Linear Discriminant Analysis) was efficient, selecting only one variable in the NIR spectral range of measures, correctly classifying all samples of the test set. When evaluating the accuracy of the calibration models SPA-MLR (Sucessive Projections Algorithm- Multiple Linear Regression) and PLS (Partial Least Square) using the elliptical confidence region it is perceived that they contain the ideal point, when the technique used was the external validation, it allows us to infer, these models lack of significant systematic errors. By analyzing these models using the cross-validation technique, we note that they do not contain the ideal point according to the elliptical region of confidence. The proposed methods are promising for determining phenotypic characteristics in a nondestructively way in castor bean genotypes.