'Marfim pattern’ of wood species in the Southern Brazilian timber market: colorimetry and visible spectra

Abstract

ABSTRACT Evaluating wood with similar characteristics is a challenging task. Taking into consideration the need for an efficient and rapid technique, the use of visible spectrophotometry to discriminate wood species is proposed. This study aimed to differentiate nine wood species with a similar yellowish color using visible spectrophotometry, associated with two classification models, to verify predictions. Color data were obtained from different anatomical sections of the wood, along with reflectance spectra in the range of 360–740 nm. Two machine learning approaches were applied: artificial neural network (ANN) and k-nearest neighbors (k-NN). In general, the mean spectra of the samples shared similar information, as they are materials of lignocellulosic origin. However, in all tests, the models built using the ANN algorithm demonstrated an accuracy in discriminating species exceeding 97%. Notably, when applying ANN to the spectra obtained solely from the longitudinal surface, superior results were obtained. Discriminative patterns can be obtained by visible spectra independent of the anatomical section. The integration of visible spectra and ANN proved to be a suitable approach to recognize “marfim pattern” wood.