Determination of percentage of mealy kernels (mitadines) in durum wheat with the Inframatic 9500 NIR instrument

Peter Åberg (1), Gautier Baudon (2), Hervé Widloecher (3), and Simon Sverkel (1)

(1) Perten Instruments, Sweden
(2) Atlantique Analyses SAS, France
(3) Perten Instruments, France


The percentage of mealy or starchy kernels in durum wheat is an important quality factor for the pasta industry. Durum kernels should ideally be glassy and non-mealy to obtain a high semolina yield. Durum lots with more than for example 20% mealy kernels may be rejected as unsuitable for semolina production and carry a price penalty.

The reference method for mealiness is laborious as it involves manual or semi-manual cutting of up to 600 kernels in half, followed by visual inspection of the cut kernel. NIR is well established in the grain industry and therefore an attractive and rapid alternative.

Figure. From left to right; Perten Inframatic 9500. Pohl farinator for cutting kernels (Bipea Ref method 204-1104) . Cut kernels showing mealy (starchy) and vitreous (glassy) kernels. 

Materials and Methods

A total of 230 whole grain durum samples from various regions of France were analyzed at Atlantique Analyses, La Rochelle, France. Mealiness was assessed using a Pohl farinator on 12 x 50 kernels. Transmission spectra (570 to 1100 nm) were measured using the Inframatic 9500, Perten Instruments, Sweden.

Traditionally, the transmission within 850 and 1050 nm is used to model the relation between spectra and constituents in cereals, but the aim here was to test also surrounding wavelengths for contribution to the model. Accordingly, PLS regressions for the limited range (850 to 1050 nm) and the whole range were developed. However, the signal-to-noise ratio varies with wavelength, and different spectral pre-processing methods were thus applied in different wavelength regions. Savitzky-Golay second derivatives 11 nm window followed by Detrend and SNV were applied in the 850 to 1050 nm range. The same pre-process methods was used for the other wavelengths, except that the window size was increased to 51 nm to account for the reduced signal-to-noise levels.


Figure: Pre-processed NIR spectra for the 850 and 1050 nm wavelength range (top) and the 570 to 1100 nm (bottom). The mealiness of the samples is indicated with color. 


It is indicated that the accuracy of the PLS model with the whole wavelength range is around 13% better than the PLS with the limited range. The resulting SECV 7.6 is approximately two times the error of the reference method.


The results indicate that there is a clear correlation between NIR and mealiness in durum wheat samples. The correlation can be improved by including the whole wavelength range from 570 to 1100 nm given that the pre-processing methods are adjusted to accommodate the noise levels in the low wavelength range.

Figure: Observed vs. cross-validated mealiness for the PLS with spectra between 850 and 1050 nm (left) and the PLS with spectra from 570 to 1100 nm (right). It is indicated that the extended spectral wavelength range improve the accuracy around 13%.


This poster was first published NIR 2013 – 16th International Conference on Near Infrared Spectroscopy, June 2–7, 2013 – 34280 La Grande-Motte, France.