Modelos de classificação de sistemas de produção de leite equantificação de ácidos graxospor espectroscopia NIR

• Ano de defesa: 2016
• Autor(a) principal: Milani, Marceli Pazini
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; BR; Ciência e Tecnologia dos Alimentos; UFSM; Programa de Pós-Graduação em Ciência e Tecnologia dos Alimentos
• Palavras-chave em Português: Calibração multivariada; Leite orgânico; Quimiometria; Chemometry; Multivariate calibration; Organic milk; CNPQ::CIENCIAS AGRARIAS::CIENCIA E TECNOLOGIA DE ALIMENTOS
• Link de acesso: http://repositorio.ufsm.br/handle/1/3417

It was evaluate the fatty acid profile of organic and conventional milk in differen seasons, the use of spectral data obtained in the near infrared region for classification of milk (organic vs conventional) and quantification of fatty acids. The tested rating models were: support vector machine (SVM) and independent flexible modeling by analogy class (SIMCA); and quantification: regression by partial least square (PLS) by interval (iPLS), principal component regression (PCR), regression by support vectors machine (SVR) and artificial neural networks (RNA). The milk samples used tanks producing units 135 and individual cows. The database used to develop the classification models were constructed from milk samples collected every two months, from July 2011 to May 2012, from 20 units producing organic milk and 20 conventional, located in the Southern of Brazil. The reference method the quantification of fatty acids used was gas chromatography (GC/FID). Spectral recordings were made in the range of 1100 to 2500nm in both samples of fresh milk as freeze-dried. The results did not show significant difference between the production system (organic and conventional) and the content of most fatty acids, including those identified as beneficial to consumer health. The use of near infrared spectroscopy associated with chemometric classification models made it possible to differentiate between milk samples from different systems, with an accuracy of 80,93% for freeze-dried samples and 59,32% for in natura samples by MVS method. Quantification of fatty acid in milk from NIR data is possible, and the selection of wavelengths by iPLS improved prediction model regarding the use of the entire spectrum. The artificial neural network has better performance than other models. The water content interfere with the performance of the models, both as to rate the samples to quantify fatty acids with significantly superior results with lyophilized samples.