Near-Infrared Hyperspectral Imaging in Tandem with Machine Learning Techniques to Identify the Near Geographical Origins of Barley Seeds

Abstract

The nondestructive identification of the geographical origins of the seeds is a crucial step in the food industry. The seeds from near geographical origins are challenging to be separated due to identical climatic and agronomic conditions. The current study implemented the idea of combining near-infrared hyperspectral imaging (NIR-HSI) with machine learning to distinguish barley seeds concerning their geographical origins. Hyperspectral images of barley seeds from four near geographical origins were captured within the range of 900–1700 nm. Sample-wise spectra were extracted from the hyperspectral images and pretreated with different spectral preprocessing techniques, viz., standard normal variate (SNV), multiplicative scatter correction (MSC), Savitzky–Golay smoothing (SGS), Savitzky–Golay first derivative (SG1), Savitzky–Golay second derivative (SG2), and detrending. Unprocessed and preprocessed sample-wise spectra were given as input to four different machine learning models. Support vector machines (SVMs), K-nearest neighbors (KNNs), random forest (RF), and partial least squares discriminant analysis (PLS-DA) were used for the classification based on the 1D spectral features. Among these classifiers, SVM showed the best classification accuracy of 93.66% when applied with the SG2 preprocessing technique. The results revealed the significance of using hyperspectral and machine learning to make a clear, fast, and accurate difference among barley varieties based on their geographical origins.