Caffeine Content
Caffeine is one of the most regulated parameters in tea for export labelling, particularly for health-oriented and decaffeinated product lines. A rapid, non-destructive screening tool enables blenders to verify caffeine concentration at intake without waiting for HPLC laboratory results. We have delivered a lite model capable of stratifying samples into compliance bands, with a full quantitative model in the development queue.
Findings
What We've Established
Caffeine Absorption Features in NIR
Caffeine's imidazole and carbonyl functional groups produce distinguishable NIR absorption features, particularly in the combination band region (4,000–5,000 cm⁻¹). These features are partially overlapping with other nitrogenous compounds in tea but are resolvable with multivariate deconvolution.
Lite Model Performance
The lite model was trained to classify samples into low, medium, and high caffeine bands corresponding to standard grade ranges for Sri Lankan black tea. Classification accuracy on the validation set was sufficient for blend-level screening and label compliance workflows, though not for precise quantification.
Methodology
Technical Approach
The lite model was trained using a reduced calibration dataset relative to the full quantitative models - sufficient for band classification but not continuous prediction. Reference values were obtained by UV spectrophotometry at 274 nm, a widely used rapid reference method for caffeine in tea. The full model will require HPLC reference values and a larger, more diverse calibration set.
Status
Where We Stand
The lite model is delivered and available for client use for blend screening purposes. The full quantitative model has been scoped but is pending dataset expansion. Priority is lower than the TF/TR fractionation models given existing client needs.
Roadmap
Next Steps
Expand calibration dataset with HPLC-referenced values for a full quantitative model
Assess feasibility of simultaneous caffeine and theanine prediction from a single scan
Evaluate application to decaffeinated tea authentication
Explore More
Related Projects
Moisture Content - Green Tea
Real-time MC measurement for green tea. Model trained on TRI-standard data and independently validated in December 2025.
View project Model ReadyTotal Polyphenols (TPP)
Total polyphenol profiling for black tea - the leading proxy for antioxidant quality and a primary auction pricing signal.
View project Model ReadySugar in Tea
Total sugar quantification across black tea grades - a key differentiator for premium blend specification and export documentation.
View projectR&D Partnerships
Interested in This Research?
If you have relevant data, domain expertise, or a measurement problem in this area, we're open to research collaboration and data-sharing agreements.