Assess whether Raman spectroscopy combined with machine learning can estimate multiple maple syrup quality attributes from one analytical workflow.
Approach:
Samples and measurements: Researchers analyzed 721 untreated maple syrup samples spanning four color grades and multiple flavor categories using a 785 nm Raman system. They collected 30 three-second spectra per sample.
Modeling: The team used principal component analysis and convolutional neural networks to examine color and flavor categories, and a random forest model trained on aqueous sucrose standards to estimate sucrose-equivalent values.
Key Findings:
Color-classification accuracy was 91% for golden and 92% for very dark syrup, compared with 68% for amber and 73% for dark syrup; errors were often between neighboring grades.
Flavor classification correctly identified 83% of samples without detectable off-flavors and 78% with the late-season “buddy” defect. Accuracy was lower for several other defect categories.
Raman fluorescence-background intensity increased with darker syrup and was inversely correlated with light transmittance; color-grade estimation accuracy ranged from 81% to 94%.
The sucrose-standard model had an R² of 0.87 and a root mean squared error of 2.2%, but syrup results were sucrose-equivalent estimates, not a validated measurement of total sugar concentration.
Interpretation:
The study reports that Raman spectroscopy and machine learning can extract information relevant to several syrup quality attributes from one dataset. The authors describe the approach as a potential rapid screening method, with uncertain samples referred for conventional testing.
Limitations:
Color grades have continuous light-transmittance boundaries, and neighboring grades were frequently confused.
Some flavor categories had overlapping or chemically diverse sensory characteristics, and rare defects were limited in the sample set.
The sucrose-equivalent model was not validated as a measurement of sugar concentration in maple syrup.
External validation across production seasons, instruments, and additional examples of rare flavor defects is needed before industrial deployment.
Conclusion:
The findings support further evaluation of Raman spectroscopy as a multi-attribute screening approach for maple syrup.
