Varietal Classification of Wheat Seeds Using Hyperspectral Imaging Technique and Machine Learning Models

Classification, recognition, and authentication of wheat grain varieties are essential because their high purity results in high yield and quality guarantee. In the present study, sixteen (16) wheat varieties harvested from the Punjab region were chosen for classification. The images of the wheat seeds were captured from both sides using a near-infrared hyperspectral imaging system that covers all the spectral bands from 900–1700 nm wavelength. Two machine learning models, support vector machine (SVM) and linear discriminant analysis (LDA), were implemented to classify wheat varieties. The models were trained separately on raw spectral data and preprocessed spectral data. Three preprocessing techniques pretreated the mean spectra: standard normal variate (SNV), Multiplicative Scatter Correction (MSC), and Savitzky-Golay Smoothing (SG Smoothing) to abolish the interference caused by instrumental and environmental factors. The support vector machine obtained the best result on the raw spectral data with a test accuracy of 93%.