NIR Instruments and Prediction Methods for Rapid Access to Grain Protein Content in Multiple Cereals

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Food Science

Background:

• Global nutrition and wellbeing depend on high-quality diets.
• Accurate food-source quality information is vital for agri-food systems.
• Traditional lab methods and benchtop NIR spectrometers are unsuitable for decentralized chains in emerging economies.

Purpose of the Study:

• To explore benchtop and portable Near-Infrared (NIR) spectroscopy instruments for grain quality analysis.
• To evaluate classical, machine learning (ML), and convolutional neural network (CNN) methods for predicting grain protein content.
• To assess the suitability of NIR spectroscopy for stakeholders in decentralized agri-food chains.

Main Methods:

• Generated NIR spectra for 328 grain samples (finger millet, foxtail millet, maize, pearl millet, sorghum) using benchtop (DS2500) and portable (HL-EVT5) NIR spectrometers.
• Developed calibration models using classical deterministic methods, ML-driven methods, and a CNN-based method.
• Validated prediction accuracy using R², Root Mean Square Error (RMSE), and Residual Predictive Deviation (RPD) metrics.

Main Results:

• All tested methods successfully built relevant calibrations for predicting grain protein (R² ≥ 0.90, RMSE ≤ 0.91, RPD ≥ 3.08).
• ML-integrated calibration methods generally improved prediction capacity.
• The portable NIR instrument (HL-EVT5) provided highly relevant quantitative protein predictions (R² = 0.91, RMSE = 0.97, RPD = 3.48).

Conclusions:

• Portable NIR spectroscopy, combined with advanced calibration methods, offers a viable solution for on-site grain quality assessment.
• Findings support the expanded use of NIR spectroscopy in agricultural research, development, and trade, particularly in emerging economies.
• This technology can contribute to achieving global goals for sustainable nutrition and food security.