فهرست نشریات علمی دسترسی آزاد ایران

Coffee species identification is essential for quality control and fraud prevention in the coffee industry. This study investigates the use of Fourier transform infrared (FT-IR) and near-infrared (NIR) spectroscopy coupled with chemometric techniques to differentiate between Arabica, Canephora, and Liberica ground coffee species. Various preprocessing methods were applied to optimize spectral data prior to model establishment. Spectral data were analyzed using principal component analysis (PCA) for visualization, followed by linear discriminant analysis (LDA) and support vector machine (SVM) for classification. The results demonstrated that LDA offered superior robustness compared to SVM. SVM was found to be highly susceptible to physical light scattering, requiring standard normal variate (SNV) correction to improve accuracy from <61 to 100%. In contrast, LDA consistently maintained high stability across various preprocessing techniques. External validation revealed distinct optimal strategies for each instrument: FT-IR models required detrending to compensate for baseline drifts, whereas NIR models favored minimal preprocessing (raw/smoothing) to achieve 100% sensitivity and specificity. Furthermore, analysis of spectral loadings identified lipid content (C–H stretching) and caffeine/protein bands as the primary chemical discriminants driving the separation. This study establishes that robust linear models (LDA) coupled with instrument-specific preprocessing constitute the optimal protocol for routine, high-throughput coffee species authentication.