Classification of different gluten wheat varieties based on hyperspectral preprocessing, feature screening, and machine learning

Classifying different gluten wheat varieties can meet diversified food needs. To rapidly classify wheat gluten types using hyperspectral data, four preprocessing methods combined with two feature screening methods and four machine learning algorithms were used in this study. Our findings indicated that the feature wavelengths extracted by ReliefF showed better classification model accuracy than that of the full wavelength classification model and the minimum redundancy maximum relevance (mRMR) classification model. The classification accuracy of continuous wavelet transform (CWT) was higher than that of original reflectance, continuous removal, and first derivative. The performance of the four classifiers were in the order support vector machine (SVM) > convolutional neural network (CNN) > random forest (RF) > K-nearest neighbor (KNN). ReliefF-CWT-SVM was identified as the optimal classification model (overall accuracy = 94.5 %). The developed combination method supplies theoretical and technical support to classify wheat varieties with different types of gluten. • Data preprocessing and feature filtering methods were compared • Machine learning models were used to classify different gluten wheat varieties • The classification accuracy of continuous wavelet transform was the highest • Continuous wavelet transform sensitive bands were 350–997 nm and 1918–2279 nm • ReliefF-continuous wavelet transform-support vector machine was the best model