Artificial neural networks and multivariate models to distinguish native and Eucalyptus charcoal based on NIR spectroscopy and x-ray fluorescence

Abstract

Charcoal is an important source of energy in Braziland can be sourced from native or planted wood. The use of wood from native forests for this purpose is prohibited in many states of the country as they may not have been managed sustainably.There is a need to develop fast and efficient methods for distinguishing charcoal from native woods and Eucalyptus (used in reforestation), and thus curbing illegal trade. The general objective of this study was to distinguish charcoal from native and Eucalyptus woods by multivariate models based onnear infrared (NIR) spectral signatures and artificial neural network based on the percentage of mineral elements.Charcoal produced at different temperatures (300, 400, 500, 600 and 700°C) and carbonization furnaces from native woods (Apuleia sp., Cedrela sp., Aspidosperma sp., Jacaranda sp., Peltogyne sp., Dipteryx sp. and Gochnatia sp.) and woods of Eucalyptus sp. from commercial forest plantations were investigated. Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA) based on NIR spectra collected on the radial face of the wood and charcoal samples were performed to identify the timber species, the origin of charcoal and the carbonization temperature. The composition and proportion of the mineral elements present in the charcoal were determined by X-ray fluorescence. Artificial neural networks (ANNs) were trained based on the mineral composition of the charcoal to predict their origin. The PLS-DA wood models made from untreated NIR spectra showed a large percentage of correct classifications (86 to 100%) for native species, except for Eucalyptus samples that were confused between the two varieties. The graph of the PCA scores revealed that the spectral similarity is greater among the charcoal produced at the same temperature than among the same species, which demonstrates the importance of this process effect. PLS-DA models based on NIR spectra were efficient in predicting carbonization temperatures, correctly classifying 87% of the samples and in predicting the native origin or charcoal Eucalyptus, especially when the origin classification was made as a function of the carbon temperature. carbonization. RNAs based on mineral element content showed potential for prediction of species and native origin or Eucalyptus, correctly classifying 74.5% and 97.9% of samples from the test batch, respectively. All techniques used in the study have potential for use in enforcement actions.