In the coffee industry, verifying a bean’s geographic origin is essential for ensuring quality and authenticity. A recent study published in Food Innovation and Advances introduces new methods for authenticating coffee origins, supporting the industry’s commitment to quality and traceability.
Image Credit: Paulius Beinaravicius/Shutterstock.com
Vibrational spectroscopy, traditionally valued in pharmaceutical and forensic science, is now gaining ground in agriculture for quality control and origin verification of biological products.
Techniques such as near-infrared (NIR), mid-infrared (FTIR), Raman, and hyperspectral imaging (HSI) spectroscopy offer rapid, non-destructive analysis of food items. However, variations in sample properties, like particle size and density, can introduce noise into spectral data, affecting accuracy. To mitigate these issues, preprocessing spectral data is essential to remove physical artifacts and improve model reliability.
In this study, four vibrational spectroscopy techniques were compared—dispersive near-infrared (DG-NIR), near-infrared hyperspectral imaging (HSI-NIR), attenuated total reflectance Fourier transform infrared (ATR-FTIR), and Raman spectroscopy—by applying various preprocessing methods to classify coffee samples from Indonesia, Ethiopia, Brazil, and Rwanda.
This preliminary research focused on identifying essential preprocessing methods and detecting potential outliers. The primary challenges involved three specific spectral data issues: offsets, slopes, and curvature, each impacting signal accuracy.
Offsets, often resulting from instrumental drift or uneven particle grinding, were absent in the data. However, slopes—particularly noticeable in Raman spectra due to fluorescence interference—and curvature in DG-NIR and HSI-NIR spectra caused by light scattering were observed. These nonlinearities, stemming from differences in sample surface properties, were addressed through targeted preprocessing techniques.
To tackle these challenges, the spectra were mean-centered before further analysis. No outliers were detected in any of the datasets, as confirmed by high KNN distances and reduced Hotelling’s T2 and Q residuals tests, all within the 95 % confidence interval.
The study emphasizes that preprocessing techniques, including normalization, scatter correction, and spectral derivations, are crucial for removing physical artifacts. Additionally, Matthew’s Correlation Coefficient (MCC) was applied as a primary decision metric to handle data imbalances, offering a more thorough model performance evaluation compared to accuracy or F1 scores.
This approach enabled the identification of optimal preprocessing treatments for each instrument, enhancing the accuracy of coffee origin classification across various countries.
Our study introduces a systematic approach to selecting the best preprocessing method, addressing a critical challenge in vibrational spectroscopy. This work not only enhances classification accuracy but also provides a robust framework for future applications in food traceability.
Dr. Joy Sim, Formulation Technologist, Danone
By demonstrating the potential of vibrational spectroscopy as a potent instrument for guaranteeing food safety and quality, with numerous applications in agriculture and beyond, this study opens the door for more effective and sustainable ways to confirm the origin of coffee and other biological materials.
Journal Reference:
Sim, J. et. al. (2024) Optimization of vibrational spectroscopy instruments and pre-processing for classification problems across various decision parameters Food Innovation and Advances. doi.org/10.48130/fia-0024-0004