Measurement of moisture content for rough rice by visible and near-infrared (NIR) spectroscopy

Abstract

The measurement of moisture content (MC) for rough rice is important in agro-industry. As the price goes higher of rice products there is a need to apply non-destructive method to measure MC. And by measuring single kernel of rice sample, we can quickly obtain the MC distribution of rice samples. This study aims to apply two spectrometers (meter A and meter B) in visible and near infrared (NIR) regions to record spectrum on three rice types, i.e. single kernel (SK), multi kernels (MK), and cracked multi kernels (CMK). Taikeng No.9 medium rough rice were randomly collected from field after harvested and they were conditioned by oven to six MC levels ranging from 11.5 to 28.7%. Two calibration methods, multiple linear regressions (MLR) and partial least square regression (PLSR), were tested to calibrate mathematical model on reflectance and absorption spectra with the corresponding MC values. Among 72 tested models, the best model was found at PLSR model with first differential of 21 gap points, which had rc = 0.97, SEC = 1.30 for calibration and rp = 0.92, SEP = 2.51 for validation. Results also suggested that ten loadings used in PLSR method could have the highest rp and lowest SEP. When three MC levels were calibrated for SK sample type, a higher rp value 0.95 was reached. The top five wavelengths selected by MLR were 905, 936, 925, 1015, and 499 nm in 400–1050 nm (meter A) and shifted to 900, 860, 970, 1610, and 830 nm when 400–2498 nm (meter B) was applied. Removing outlier also could improve the model both in six MC groups and three MC group models.