Determination of moisture in raw coffee by near infra-red reflectance spectroscopy and multivariate regression

Near infra-red reflectance (NIR) spectroscopy was used to measure the moisture content in raw coffee. Different models using partial least squares (PLS) with data pre-processing were used. Regression models were built with 157 spectra of the samples of raw coffee collected using a near infrared spectrometer with an accessory of diffuse reflectance, between 4500 and 10000 cm(-1). The original NIR spectra went through different transformations and mathematical pre treatments, such as the Kubelka-Munk transformation; multiplicative signal correction (MSC); spline smoothing and movable average, and the data were scaled by variance. The regression model permitted the determination of the moisture content of the raw coffee samples with a standard error of calibration (SEC) = 0.569 g.100 g(-1); standard error of validation = 0.298 g.100 g(-1); correlation coefficient (r) 0.712 and 0.818 for calibration and validation, respectively, and average relative error of 4.1% for validation samples.