Diferenciação de espécies de Eucalipto usando análise multivariada e espectroscopia no infravermelho

Abstract

Eucalyptus wood is used for the production of cellulose, furniture, civil construction, among other applications, with the most suitable species for each use. The separation and classification of wood species is based on characteristics such as color, smell, taste, texture or brightness. Alternatively, more elaborate laboratory analyzes can be used, demanding more time and cost. Therefore, it is necessary to develop a new accurate, fast and low-cost method for classifying wood species. In this work, we investigated the use of Fourier transform infrared spectroscopy (FTIR) associated with machine learning as an alternative for the classification of 6 different tree species of the Eucalyptus family: Eucalyptus camaldulensis, Corymbia citriodora, GG100, Eucalyptus grandis, Eucalyptus saligna and Eucalyptus urophylla. Using machine learning algorithms on data obtained from principal component analysis (PCA) of the FTIR spectrum of powdered sapwood samples, it was possible to obtain predictive models for species classification. The infrared spectrum was divided into three regions, 4000 – 700, 3000 – 2800 and 2000 – 700 cm-1, to analyze the influence of cellulose, hemicellulose, lignin and plant extractives on the different classifiers. The Support Vector Machine (SVM) algorithm achieved an accuracy greater than 90% - in all studied intervals - in the Leave One Out Cross Validation tests.

Keywords: Eucalyptus; FTIR; Machine Learning; Multivariate Analysis: Sap-wood.