Brown mustard (Brassica juncea (L.) is an important oilseed crop that is mostly used to produce edible oils, industrial oils, modified lipids and biofuels in subtropical nations. Due to its higher level of commercial use, the species has a huge array of varieties/cultivars. The purpose of this study is to evaluate the use of visible near-infrared (Vis-NIR) spectroscopy in combination with multiple chemometric approaches for distinguishing four B. juncea varieties in Korea. The spectra from the leaves of four different growth stages of four B. juncea varieties were measured in the Vis-NIR range of 325-1075 nm with a stepping of 1.5 nm in reflectance mode. For effective discrimination, the spectral data were preprocessed using three distinct approaches, and eight different chemometric analyses were utilized. After the detection of outliers, the samples were split into two groups, one serving as a calibration set and the other as a validation set. When numerous preprocessing and chemometric approaches were applied for discriminating, the combination of standard normal variate and deep learning had the highest classification accuracy in all the growth stages achieved up to 100%. Similarly, few other chemometrics also yielded 100% classification accuracy, namely, support vector machine, generalized linear model, and the random forest. Of all the chemometric preprocessing methods, Savitzky-Golay filter smoothing provided the best and most convincing discrimination. The findings imply that chemometric methods combined with handheld Vis-NIR spectroscopy can be utilized as an efficient tool for differentiating B. juncea varieties in the field in all the growth stages.
Keywords: Brassica juncea; deep learning; machine learning; spectroscopy; variety discrimination; visible near-infrared.