The combination of near infrared spectroscopy (NIR) and multivariate calibration using the partial least square - PLS method for the determination of the total protein level in raw coffee samples was investigated. The total protein levels were initially determined using the Kjeldhal method as the reference method. Regression models were built from the spectra in the NIR region of the raw coffee samples. Spectra of 159 samples were recorded, using an accessory of diffuse reflectance, in the range of 4500 and 10000cm-1 with 4cm-1 resolution. To the raw spectral data, different transformations and mathematical pretreatment such as Kubelka-Munk transformation; multiplicative sign correction (MSC); spline smoothing; first derivative and boxcar averaging were applied. The data was also preprocessed by scaling each column by its variance. The analytical method proposed is non invasive and the cost and time of analysis are very much reduced, making possible the fast and direct determination of the total protein content. The prediction error of protein levels given by the regression model were in average lower than 6.7%, indicating that this method is a good alternative for protein determination in raw coffee samples.
protein; near infrared spectroscopy; coffee; PLS
