Determinação da proteína, óleo e umidade por espectroscopia NIR em grãos de soja do Estado de Roraima

Abstract

In this study the potentialities of (NIR – Near Infrared) spectroscopy and chemometrics techniques are combined aiming to determination of protein, oil and moisture in whole-grain soybean cultivated in the experimental field of the Roraima Embrapa, for which it was used as reference the Kjeldahl method, the extraction method by Soxhlet and the oven method, respectively. In the development of the models were studied several pre-processing types of the spectra data, spectral range and sample set. The regression method used in the development of the models was PLS (Partial Least Square) using mean-centered data, for a set of 90 spectra (calibration and cross validation). The evaluation of the models was made observing the values of correlation coefficient (R) and the root mean square error. The prevision values were estimated for the other 45 spectra nor used in the calibration set. For the determination of protein level, best results were obtained using Multiplicative Signal Correction (MSC) in the 1510-1980 nm range, obtaining RMSEP with values 0.66. For the determination of oil level, best results were obtained using MSC with transform Base Line (BL) in the 1110-2500 nm range, obtaining RMSEP with values 0.40. And for the determination of moisture level, best results were obtained using Standard Normal Variate Transformation (SNV) with BL, in the 1870-2020 nm and 1400-1500 nm combined range, obtaining RMSEP with values 0.12. The medium relative errors among predicted values and the measured values were inferior to 2%. With these results conclude that good models aiming the prediction of protein, oil and moisture levels in the whole-grain soybean from Roraima can be obtained. And that the technique NIR allied to the PLS regression is adequate to allow besides fast and precise determinations, the not destruction of the samples nor the use of reagents use and for not create wastes which are harmful to the environment.