Raman Spectroscopy Takes On Maple Syrup Grading

Machine learning extracts information on color, flavor defects, and sugar content from the same spectral dataset

1

Researchers analyzed 721 untreated maple syrup samples with 785-nm Raman spectroscopy, collecting 21,630 spectra for machine-learning analysis.
2

Convolutional neural networks classified golden and very dark syrups with 91% and 92% accuracy, respectively, but accuracy was lower for amber and dark grades.
3

Flavor-classification accuracy was 83% for syrups without detectable defects and 78% for syrups with the late-season “buddy” defect.
4

Raman fluorescence background intensity increased as syrup darkened and supported color-grade estimates with 81% to 94% accuracy.
5

A sucrose-standard model yielded an R² of 0.87 and a 2.2% root mean squared error, but syrup estimates were sucrose equivalents, not validated sugar concentrations.

1