Rapid and non-destructive monitoring of the drying process of glutinous rice using visible-near infrared hyperspectral imaging

Rapid and non-invasive monitoring of the drying process of glutinous rice is crucial to ensure the effective production of desired dried grain. In this study, visible-near infrared hyperspectral imaging coupled with computational intelligence was used to detect the variation in moisture content (MC), change in colour ( Δ E ), and golden index (GI) of glutinous rice during drying. Different preprocessing methods and effective wavelength selection techniques were used to eliminate the noise and redundant wavelength in the reflectance spectra, and predictive models were developed for the glutinous rice quality. Savitzky-Golay first derivative (SG1D) showed the best preprocessing performance ( 0.9564 ≤ R P 2 ≤ 0.9781 , 0.0177 ≤ R M S E P ≤ 0.8242 and 1.28 ≤ M A P D ≤ 5.90 for PLSR model ). The best performance accuracy ( R P 2 ≥ 99.99 % ) was obtained when the SG1D and Gaussian process regression (GPR) model were combined with iteratively retained informative variable algorithm (SG1D-IRIV-GPR), variable iterative space shrinkage (SG1D-VISSA-GPR) and variable combination population analysis (SG1D-VCPA-GPR) for the prediction of MC, GI, and Δ E , respectively. The study showed that visible-near infrared hyperspectral imaging coupled with computational intelligence can be used to monitor the quality of glutinous rice during the drying process.