Discrimination of “Louros” wood from the Brazilian Amazon by near-infrared spectroscopy and machine learning techniques

Abstract

The integration of near infrared spectroscopy (NIR) with machine learning techniques can be an adequate method for discrimination of wood species with commercial value. The aim of this study was to discriminate wood samples marketed as “Louros” from the Brazilian Amazon based on near-infrared spectroscopy and machine learning techniques. Samples of louro vermelho, louro branco, louro pimenta, louro preto, louro rosa, itauba, itauba amarela and preciosa were collected by members of two extractivist communities, Paraiso and Arimum, located in the “Green Forever” Extractivist Reserve in Para state. Near-infrared spectra were obtained in the range 4000–10,000 $$\hbox {cm}^{-1}$$ , with resolution of $$4 \hbox { cm}^{-1}$$ , directly from sample surfaces oriented in the three anatomical sections: transverse, radial and tangential. This work tests three machine learning approaches—namely support vector machine (SVM), partial least squares-discriminant analysis (PLS-DA), and k-Nearest Neighbors (k-NN). The repeated k-fold cross validation method based on stratification and blocking was used to estimate the performance of the machine learning models. To build learning models, based on near infrared spectra, two situations were considered: (1) applying spectra from all wood sections and (2) using only spectra from one wood section. In general, mean spectra of “Louros” samples were similar. In all tests, models built with PLS-DA algorithm had accuracy and F1-Score superior to 97%. When analyzing PLS-DA applying spectra from only one wood section, tangential section had results slightly superior. Discriminative patterns can be obtained by near infrared spectra independent of anatomical section. The integration from NIR and PLS-DA was an adequate approach to recognize wood from “Louros” group.