Prediction of the basic density of tropical woods by near-infrared spectroscopy

Abstract

Background: Determining the basic density of the wood is usually defined as a simple process, but it requires caution and the operator's skill to avoid errors in the analysis. In addition, it involves sample preparation and time to saturate the wood until obtaining the dry sample mass. The development of alternative measurement techniques could reduce the time to obtain the results and provide reliable values. Therefore, this study aimed to develop multivariate models to estimate the basic density of native woods using near-infrared spectra (NIR). Basic densities were determined by the water immersion method, and the values were associated with NIR signatures. Spectra were directly collected on the wood transversal and radial faces with an integrating sphere. Partial least squares regression (PLS-R) was calibrated and validated to estimate basic density based on spectral signatures. Results:In the cross-validation and prediction of the models, the results were promising. The coefficients of determination varied from 0.87 to 0.93 with a standard error of 0.01 %. The partial least squares discriminant analysis (PLS-DA) efficiently classified the wood species. The ratio of performance to deviation obtained satisfactory values, a minimum of 2.81 and a maximum of 4.20. Conclusion:The statistical parameters of the models based on NIR spectra showed potential for density measurements in floors, furniture, and solid wood products.

Impact of surface quality on the identification of tropical wood species using benchtop and portable NIR instruments

“…The rst model was tted with the original data, while the other ve models were tted with the following pre-treatments: normalization, rst (1d) and second order (2d) derivatives, multiplicative dispersion correction (MSC) and standard normal variation (SNV). The rst (13-point lter and secondorder polynomial) and second derivative (25-point lter and second-order polynomial) were based on Savitzky-Golay (SG) algorithm, as suggested by Medeiros et al (2023).…”