Key Points
- Researchers from the University of Belgrade demonstrated that combining Raman and FT-IR spectroscopy with machine learning algorithms, especially support vector machines (SVM), can accurately classify seed varieties of paprika, tomato, and lettuce, achieving up to 100% accuracy in some cases.
- Raman spectroscopy proved more sensitive than FT-IR in detecting molecular differences, and combining both methods further improved classification robustness while offering non-destructive, preparation-free, and rapid testing advantages.
- Despite promising results, challenges such as overfitting and model repeatability remain, highlighting the need for continued refinement to ensure broader applicability and long-term reliability in seed quality assessment and food fraud prevention.
A new study from the University of Belgrade has demonstrated the potential of combining Raman and Fourier-transform infrared (FT-IR) spectroscopy with advanced machine learning (ML) algorithms to accurately classify seed varieties in major vegetable crops. This study, which was published in the journal Plants, this research examined how this method can be used for crop variety discrimination and seed quality assurance (1).
What crops did the study investigate?
The researchers focused on three crops for their study: paprika (Capsicum annuum L.); tomato (Lycopersicon esculentum Mill.); and lettuce (Lactuca sativa L.). These crops were selected because these crops are highly sought after by consumers. Rather than comparing different crop species, which show macroscopic seed differences such as size and color, the research focuses on varietal differences within each species.
What was the experimental procedure?
As part of the experimental procedure, the researchers collected the spectral data from seed samples by using Raman spectroscopy and Fourier transform infrared (FT-IR) spectroscopy. Several pre-processing methods were used in the study, including smoothing, linear baseline correction, unit vector normalization, multiplicative scatter correction, and second-order derivatives (1). After pre-processing, principal component analysis (PCA) was employed to reduce dimensionality, followed by the application of various classification algorithms. The classification algorithms used were support vector machines (SVM), partial least squares discriminant analysis (PLS-DA), and PCA-quadratic discriminant analysis (PCA-QDA) (1).
Which model performed the best?
Out of all the models tested in the study, SVM consistently outperformed the others in classification accuracy. The researchers found that Raman spectroscopy paired with SVM achieved classification rates of 100.00% for lettuce, 99.37% for paprika, and 92.71% for tomato (1). FT-IR spectroscopy with SVM followed closely behind, with accuracy rates of 99.37% for lettuce, 92.50% for paprika, and 97.50% for tomato (1). When data from both Raman and FT-IR spectra were merged, classification accuracy improved even further, reaching 100.00% for lettuce and tomato, and 95.00% for paprika (1).
