Raman and FT-IR Spectroscopy Coupled with Machine Learning for the Discrimination of Different Vegetable Crop Seed Varieties

The aim of this research is to investigate the potential of Raman and FT-IR spectroscopy as well as mathematical linear and non-linear models as a tool for the discrimination of different seed varieties of paprika, tomato, and lettuce species. After visual inspection of spectra, pre-processing was applied in the following combinations: (1) smoothing + linear baseline correction + unit vector normalization; (2) smoothing + linear baseline correction + unit vector normalization + full multiplicative scatter correction; (3) smoothing + baseline correction + unit vector normalization + second-order derivative. Pre-processing was followed by Principal Component Analysis (PCA), and several classification methods were applied after that: the Support Vector Machines (SVM) algorithm, Partial Least Square Discriminant Analysis (PLS-DA), and Principal Component Analysis-Quadratic Discriminant Analysis (PCA-QDA). SVM showed the best classification power in both Raman (100.00, 99.37, and 92.71% for lettuce, paprika, and tomato varieties, respectively) and FT-IR spectroscopy (99.37, 92.50, and 97.50% for lettuce, paprika, and tomato varieties, respectively). Moreover, our novel approach of merging Raman and FT-IR spectra significantly contributed to the accuracy of some models, giving results of 100.00, 100.00, and 95.00% for lettuce, tomato, and paprika varieties, respectively. Our results indicate that Raman and FT-IR spectroscopy coupled with machine learning could be a promising tool for the rapid and rational evaluation and management of genetic resources in ex situ and in situ seed collections.