Hyperspectral imaging for detection of macronutrients retained in glutinous rice under different drying conditions

This study detected the macronutrients retained in glutinous rice (GR) under different drying conditions by innovatively applying visible-near infrared hyperspectral imaging coupled with different spectra preprocessing and effective wavelength selection techniques (EWs). Subsequently, predictive models were developed based on processed spectra for the detection of the macronutrients, which include protein content (PC), moisture content (MC), fat content (FC), and ash content (AC). The result shows the raw spectra-based model had a prediction accuracy ( $$ ) of 0.6493, 0.9521, 0.4594, and 0.9773 for PC, MC, FC, and AC, respectively. Applying Savitzky Golay first derivatives (SG1D) method increases the $$ value to 0.9972, 0.9970, 0.9857 and 0.9972 for PC, MC, FC, and AC, respectively. Using the variable iterative space shrinkage algorithm (VISSA) as EWs reduces the spectral bands by over 60%, and this increases the accuracy of the model (SG1D-VISSA-PLSR) to 100%. Therefore, the developed SGID-VISSA-PLSR can be used to build a smart and reliable spectral system for detecting the macronutrients in GR grains.