Mid-Level data fusion techniques for gum rosin discrimination according to Pinus species, tapping methods, and geographic origins
Résumé
Gum rosin, an essential industrial resource, is primarily derived from pine species, with significant demand across adhesives, coatings, and related industries. With the current growth trajectory of the global rosin market, ensuring product authenticity and quality is fundamental. This study focuses on authenticating gum rosin, specifically that extracted from P. pinaster in Southwest France using the BioGemme method, which is recognized for its purity due to a closed-cup tapping approach. Sixty-one gum rosin samples, representing three Pinus species, three tapping methods, and four geographical origins, were analyzed using a multi-method approach: Gas Chromatography-Flame Ionization Detection (GC-FID), standardized quality assessments, and Near-Infrared (NIR) spectroscopy. To achieve robust sample differentiation, Partial Least Squares-Discriminant Analysis (PLS-DA) was applied on individual blocks, as well as mid-level data fusion methods (SO-PLS-LDA and SO-CovSel-LDA), to combine the chemical composition, quality attributes, and NIR spectral fingerprints. Results reveal that data fusion significantly enhances the discriminative capacity, particularly with the SO-PLS-LDA model, reaching a cross-validated MCC of 0.98. Key chemical markers were identified, including diterpenes and resin acids, along with acid number from quality attributes and distinctive NIR fingerprints. This integrative approach underlines the added value of combining data sources for accurate gum rosin classification and highlights the potential for handheld NIR spectrometers as cost-effective tools for on-site discrimination. This work establishes a foundational methodology for standardized gum rosin discrimination and paves the way for future studies in industrial applications.