Combining near-infrared hyperspectral imaging and ANN for varietal classification of wheat seeds

Abstract

High-quality seeds have a positive effect on plant growth and crop yield. Varietal purity is one of the critical quality parameters to evaluate the quality of the seeds. Presently, the seeds are segregated manually by visual inspection, which is subjective and prone to human errors. Moreover, the seeds are crushed into powdered form for chemical analysis, indicating the need to explore non-destructive techniques. In this study, a near-infrared hyperspectral imaging (NIR-HSI) system in the spectra range of 900-1700 nm was used to classify the wheat seeds variety in a non-destructive manner. The seeds from 15 different Indian wheat varieties were included in the study. The mean reflectance spectrum was extracted from each seed and pre-treated using various spectral pre-processing 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. In addition, five different classifiers, namely artificial neural network (ANN), support vector machines (SVM), partial least squares discriminant analysis (PLS-DA), random forest (RF) and K-nearest neighbors (KNN). The results showed that the ANN model coupled with the SG2 pre-processing technique gave the best classification accuracy of 97.77%. The current research work investigated the NIR-HSI system combined with an ANN model to accurately, rapidly and non-destructively classify wheat seeds variety.