Avaliação da composição bromatologica de silagens de milho: análise em nivel de campo, resultados rápidos e rastreáveis utilizando espectroscopia de infravermelho próximo

• Ano de defesa: 2024
• Autor(a) principal: Pereira, Stela Näetzold
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Zootecnia; UFSM; Programa de Pós-Graduação em Zootecnia; Centro de Ciências Rurais
• Palavras-chave em Português: NIR portátil; Modelo de calibração; Processamento de amostras; Redes neurais; Portable Nirs; Model calibration; Neural network; Sample processing; CNPQ::CIENCIAS AGRARIAS::ZOOTECNIA
• Link de acesso: http://repositorio.ufsm.br/handle/1/31729

The present study aimed to produce multivariate models to determine the parameters of the chemical composition of corn silages, and evaluate the effect of different ways of processing samples and the type of portable devices, from 900 to 1700nm, which employ near-infrared reflection (NIR) on the performance of these models. In the evaluation of different forms of processing (whole fresh samples, crushed fresh samples, dried in an electric fryer/crushed and those dried in an oven/ground) of corn silage (n=370) in the parameters of an exploration model for analysis with a portable NIRS device. The estimation of the chemical composition of corn silages was not adequate when the samples were in whole, fresh form, presenting coefficients of determination (R2) lower than 0.75. While, among the forms of processing evaluated, fresh crushed samples had better prediction of dry matter (DM) content (R2val= 0.82; validation mean square error (RMSEV)= 2.59%), detergent fiber acid (R2val= 0.72; RMSEV= 3.67); starch (R2val= 0.60; RMSEV= 5.09) and the satisfactory prediction for neutral detergent fiber (NDF), acid detergent fiber (ADF) and energy (NDT, ED and EM) of the silage. The process of drying the samples in an electric fryer provided an improvement in the prediction of crude protein and ether extract contents about the fresh forms of the silages. Regarding the performance of the two devices evaluated (SN6100152 from Texas Instruments and MR-3B51R002 from InnoSpectra), the SN6100152 device showed better prediction for all tested variables, with values for DM content being observed for crushed form (R2val= 0.82; RMSEV=2.59), NDF (R2val= 0.72; RMSEV=3.67), FDA (R2val= 0.77; RMSEV=2.79) and starch (R2val= 0.60; RMSEV=5, 09) and lower performance (R2val < 0.42) for gray, PB and EE. Furthermore, the use of the deep artificial neural network technique was efficient in improving the calibration models with the standard PLS technique with MS values of R2=0.80; RMSEP=3.20 and R2=0.62; RMSEP=5.81, respectively. Therefore, portable devices are capable of evaluating the bromatological composition of silage in situ, and the SN6100152 device and the crushed processing of in natura samples using the neural network technique obtained the best prediction models.