Classification of wheat varieties by PLS-DA and LDA models and investigation of the spatial distribution of protein content using NIR spectroscopy

Abstract

Near-Infrared (NIR) spectroscopy gives information about the chemical properties of objects, and it is of particular interest for agricultural and food science applications given its rapid and non-destructive nature. This paper presents a study on the prediction of protein content in whole wheat kernels by NIR, comparing two statistical discriminant analysis techniques, namely PLS-DA and LDA, to classify wheat varieties and protein levels at the farm’s site. Data were collected from 3 varieties collected from 9 farms located in the same region, with a total of 54 samples analyzed. The NIR spectrometer used had a range of 950-1650 nm and it was used to classify different wheat samples according to their varieties and protein level. The optimal spectral pre-processing for the current application was Savitzky-Golay followed by Multiplicative Scatter Correction (SG+MSC), which resulted in R² of 0.82 and 0.79 and RMSE of 0.73 and 0.79 for the calibration and validation datasets, respectively. Among the three varieties investigated, only Gaskojhen (Gas) variety had a classification rate above 75%, while the two varieties Mih (Mihan) and Pish (Pishgam) were regarded as one class. Comparing PCA-LDA and PLS-DA, the latter showed better potential in varietal identification compared to PCA-LDA. Investigating the protein changes at different points of the farm revealed that sampling location had a significant effect on the protein content. The ability of NIR to classify different varieties indicates that NIR can be useful in assessing wheat quality, and can give helpful information in varietal identification.