Abstract
Near-infrared (NIR) spectroscopy combined with multivariate analysis has proven to be a fast and efficient method for identifying wood species. Despite significant technical advances in recent years, challenges remain that limit its application in field conditions, particularly the influence of sample surface preparation on the performance of classification models. This study aimed to evaluate the impact of wood surface quality on the performance of NIR instruments in identifying tropical wood species. Wood samples were collected from fields and log yards and prepared using different tools. NIR spectra were recorded using portable and benchtop NIR instruments on the transverse surfaces of wood specimens subjected to five treatments: (1) field conditions (untreated), (2) chainsaw, (3) circular saw, (4) bandsaw, and (5) sandpaper. Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA) were performed using the NIR signatures. Spectra collected from surfaces prepared with a circular saw and sandpaper showed clearer groupings in the PCA score plot, facilitating the identification of distinct wood species. Cross-validated PLS-DA models showed high success rates, with classification accuracies ranging from 95.3% to 99.2% for untreated, circular saw, bandsaw, and sanded surfaces. Wood surfaces prepared with a chainsaw yielded lower classification accuracies: 88.7% for benchtop and 92.8% for portable NIR sensors. These results highlight the potential of NIR spectroscopy for classifying tropical woods, even when surface quality varies.