Sorghum Grains Grading for Food, Feed, and Fuel Using NIR Spectroscopy

Abstract

Near-infrared spectroscopy (NIR) is a non-destructive, fast, and low-cost method to measure the grain quality of different cereals. However, the feasibility for determining the critical biochemicals, related to the classifications for food, feed, and fuel products are not adequately investigated. Fourier-transform (FT) NIR was applied in this study to determine the eight biochemicals in four types of sorghum samples: hulled grain flours, hull-less grain flours, whole grains, and grain flours. A total of 20 hybrids of sorghum grains were selected from the two locations in China. Followed by FT-NIR spectral and wet-chemically measured biochemical data, partial least squares regression (PLSR) was used to construct the prediction models. The results showed that sorghum grain morphology and sample format affected the prediction of biochemicals. Using NIR data of grain flours generally improved the prediction compared with the use of NIR data of whole grains. In addition, using the spectra of whole grains enabled comparable predictions, which are recommended when a non-destructive and rapid analysis is required. Compared with the hulled grain flours, hull-less grain flours allowed for improved predictions for tannin, cellulose, and hemicellulose using NIR data. This study aimed to provide a reference for the evaluation of sorghum grain biochemicals for food, feed, and fuel without destruction and complex chemical analysis.

Keywords: FT-NIR; PLSR; biochemical composition; feed; food; fuel; sorghum grains.

Figure 1

The morphological classification of sorghum grains by means of hulled and hull-less grains, also used as whole grains and flours by means of Fourier Transform near-infrared spectroscopy (FT-NIR) spectral data collection followed by reference analysis. The spectra were processed by combination of pre-treatments, cross-validated, calculation of principal components (PCs) by principal component analysis (PCA), divided into calibration and validation subsets and ultimately, partial least square regression (PLSR) models were built using FT-NIR spectral and reference data.

Figure 2

Summary of spectral absorbance and characteristic wavelengths distribution measurements for sorghum whole grains vs. flours sample set (n = 98). The range of coefficient variation (CV, blue line), mean absorbance (black line), minimum/maximum (green dotted line) spectral absorbance, total range of absorbance measurements (red line) for each wavelength (cm⁻¹), and SD+/– SD between sample sets were calculated.

Figure 3

Summary of spectral absorbance and characteristic wavelengths distribution measurements for sorghum hulled (n = 61) vs. hull-less grain flours sample set (n = 37). The CV, coefficient of variation (blue line) (SD/mean), mean absorbance (black line), minimum/maximum (green dotted line) spectral absorbance, total range of absorbance measurements (red line) for each wavelength (cm⁻¹), and +/– SD between sample sets were calculated.

Figure 4

Comparison of scatter plots of measured vs. predicted values (g kg⁻¹) of the whole grains for starch, protein, fat, tannin, cellulose, hemicellulose, lignin, and ash of sorghum grains for the external validation subsets based on PLSR models. The black and gray lines in plot showed mean, maximum, and minimum values while dotted lines showed average chemical composition between calibration and validation sample set.

Figure 5

Comparison of scatter plots of measured vs. predicted values (g kg⁻¹) of the whole grain flours for starch, protein, fat, tannin, cellulose, hemicellulose, lignin, and ash of sorghum grains for the external validation subsets based on PLSR models. The black and gray lines in plot showed mean, maximum, and minimum values while dotted lines showed average chemical composition between calibration and validation sample set.

Figure 6

Comparison of scatter plots of measured vs. predicted values (g kg⁻¹) of the hulled grain flours for starch, protein, fat, tannin, cellulose, hemicellulose, lignin, and ash of sorghum grains for the external validation subsets based on PLSR models. The black and gray lines in plot showed mean, maximum, and minimum values while dotted lines showed average chemical composition between calibration and validation sample set.

Figure 7

Comparison of scatter plots of measured vs. predicted values (g kg⁻¹) of the hull-less grain flours for starch, protein, fat, tannin, cellulose, hemicellulose, lignin, and ash of sorghum grains for the external validation subsets based on PLSR models. The black and gray lines in plot showed mean, maximum, and minimum values while dotted lines showed average chemical composition between calibration and validation sample set.