Raman spectroscopy combined with machine learning could provide several maple syrup quality measurements from a single analytical workflow, potentially reducing reliance on separate instrumental and sensory tests.
Commercial grading currently uses refractometry to measure soluble solids, spectrophotometry to determine color, and trained technicians to identify flavor defects. Chromatography or mass spectrometry may provide more detailed compositional information, but these techniques require additional time, equipment, and sample preparation.
The study assessed whether one Raman dataset could support several of these measurements. Saba Bashir, Talia Gélinas, and Jean-François Masson of the Université de Montréal, Canada, analyzed 721 maple syrup samples supplied by the Producteurs et Productrices Acéricoles du Québec.
The samples covered the golden, amber, dark, and very dark color grades, as well as syrups with no detectable defect and several categories of off-flavor. Each sample was analyzed without chemical treatment using a 785 nm Raman system. The researchers collected 30 consecutive three-second spectra per syrup, producing 21,630 spectra.
Principal component analysis revealed broad differences between color and flavor groups but could not fully separate them. The team therefore trained convolutional neural networks to identify nonlinear spectral patterns associated with each category.
Color-classification accuracy reached 91 percent for golden syrup and 92 percent for very dark syrup, falling to 68 percent for amber and 73 percent for dark samples. Most errors occurred between neighboring grades, which are defined by continuous ranges of light transmittance rather than distinct chemical boundaries.
Flavor classification varied by defect. The model correctly categorized 83 percent of syrups without off-flavors and 78 percent of samples with the late-season “buddy” defect. Accuracy was lower for several categories with overlapping or chemically diverse sensory characteristics.
The researchers also extracted information from the broad fluorescence background normally removed during Raman preprocessing. Background intensity increased as syrup became darker and correlated inversely with light transmittance, allowing color grades to be estimated with accuracies ranging from 81 to 94 percent.
A separate random forest model trained using aqueous sucrose standards produced an R² of 0.87 and a root mean squared error of 2.2 percent. However, its application to maple syrup generated sucrose-equivalent estimates rather than a fully validated measurement of sugar concentration because syrup contains additional sugars, organic acids, minerals, and phenolic compounds.
The results suggest Raman spectroscopy could eventually provide rapid screening for several quality attributes before uncertain samples are referred for conventional testing. External validation using other production seasons, instruments, and more examples of rare flavor defects will be needed before industrial deployment.
Newsletters
Receive the latest analytical science news, personalities, education, and career development – weekly to your inbox.
