Classifying rapeseed varieties using Fourier transform infrared photoacoustic spectroscopy (FTIR-PAS)

• FTIR-PAS was utilized to classify three cultivars of rapeseeds.
• PLS-DA and SVM were used to develop classification models.
• SVM was found to give better predictive accuracy.
• SPA could select relevant variables to further refine classification models.

This study proposed a methodology for classification of rapeseed varieties using Fourier transform infrared photoacoustic spectroscopy (FTIR-PAS). For this purpose, principal components analysis (PCA) was first used to reveal the separation of three varieties of rapeseeds, and then partial least squares-discriminant analysis (PLS-DA) and support vector machines (SVM) were employed for the classification task. The overall classification error rates (ERs) of prediction set were 7.5% and 0 for the models of PLS-DA and SVM, respectively. Furthermore, successive projections algorithm (SPA) was adopted to choose an appropriate variable subset as the inputs of PLS-DA and SVM. Both SPA-PLS-DA and SPA-SVM models gave improved predictive accuracy with significantly reduced model variables. The results of this study had showed the good performance of FTIR-PAS as a rapid, non-destructive and objective tool for classifying varieties of rapeseeds.