Computer vision based on GCxGC-MS/FID for Coffee origin authentication

Rationale: Computer Vision, a proven Artificial Intelligence (AI) technique, is effectively applied to authenticate coffee origin by extracting meaningful patterns from complex chromatographic data. Using comprehensive two-dimensional gas chromatography (GC×GC), we capture detailed chemical fingerprints of coffee samples. These are processed through a robust Computer Vision workflow that combines Untargeted and Targeted (UT) fingerprinting to generate composite Class Images representing distinct geographical origins (1–3). This approach enables reliable, scalable classification of coffee samples, overcoming the limitations of traditional pairwise comparison methods. Originally developed in other food applications, this workflow now delivers actionable insight into the traceability and authenticity of coffee, providing a high-resolution view of volatile organic compound (VOC) profiles with proven performance across diverse sample sets. Methods: Samples of the different origins of commercial roasted and grounded coffees were supplied by Illycaffe S.p.A (Trieste, Italy), including 32 coffees for moka preparation packed vacuum-sealed metal cans. Origins included were: Brasil, Colombia, Ethiopia, Guatemala and India. Volatilome analyses were carried out by a HS-SPMEGC×GC- MS/FID with reverse-inject differential-flow modulation platform. Automated workflow of raw signals was realized by GC-Image (GC Image LCC, Nebraska, USA). Data analysis and machine learning were performed using Python code (version 3.11.8) within a Jupyter Notebook environment. Results: Computer vision combined with chromatographic fingerprinting can effectively enable Augmented Visualization facilitating access to samples’ chemical code and interpretation of the phenomena related to it. A comparison of machine learning algorithms based on PCA-LDA, SVM, RF, PLS-DA and LR on cross-validated data were done in the ability to classify coffee origin based on chromatographic fingerprinting images.