Wood density is an important criterion for material classification, as it is directly related to quality of wood for structural use. Several studies have shown promising results for the estimation of wood density by near infrared spectroscopy. However, the optimal conditions for spectral acquisition need to be investigated in order to develop predictive models and to understand how anisotropy and surface roughness affect the statistics of predictive partial least square regression models. The aim of this study was to evaluate how the spectral acquisition technique, wood surface, and the surface quality influence the ability of partial least square-based models to estimate wood density. Near infrared spectra were recorded using an integrating sphere and fiber-optic probe on the tangential, radial, and transverse surfaces machined by circular and band saws in 278 wood specimens of six-yearold Eucalyptus hybrids. The basic density values determined by the conventional method were then correlated with near infrared spectra acquired using an integrating sphere and fiber-optic probe on the wood surfaces by means of partial least square regressions. The most promising models for predicting wood density were generated from near infrared spectra obtained from the transverse surface machined by the bandsaw, via an integrating sphere (r 2 p ¼ 0:87, RMSEP ¼ 23 kg m À3 and RPD ¼ 3.0) as well as for the optic fiber (r 2 p ¼ 0:78, RMSEP ¼ 35 kg m À3 and RPD ¼ 2.1). Surface quality affected the spectral information and robustness of predictive models with a rougher surface, caused by band sawing, showing better results.

Background: Near infrared (NIR) spectroscopy has been successfully applied to estimate the chemical, physical and mechanical properties of various biological materials, including wood. This study aimed to evaluate basic density calibrations based on NIR spectra collected from three wood faces and subject to different mathematical treatments. Methods: Diffuse reflectance NIR spectra were recorded using an integrating sphere on the transverse, radial and tangential surfaces of 278 wood specimens of Eucalyptus urophylla x Eucalyptus grandis. Basic density of the wood specimens was determined in the laboratory by the immersion method and correlated with NIR spectra by Partial Least Squares regression. Different statistical treatments were then applied to the data, including Standard Normal Variate, Multiplicative Scatter Correction, First and Second Derivatives, Normalization, Autoscale and MeanCenter transformations. Results: The predictive model based on NIR spectra measured on the transverse surface performed the best (R²cv = 0.85 and RMSE = 25.5 kg/m³) while the model developed from the NIR spectra measured on the tangential surface had the poorest performance (R²cv = 0.53 and RMSE = 46.8 kg/m³). The difference in performance between models based on original (untreated) and mathematically-treated spectra was minimal. Conclusions: Multivariate models fitted to NIR spectra were found to be efficient for predicting the basic density of Eucalyptus wood, especially when based on spectra measured on the transversal surface. For this data set, models based on the original spectra and mathematically treated spectra had similar performance. The reported findings show that mathematical transformations are not always able to extract more information from the spectra in the NIR.

No abstract

This study investigated the feasibility of using near infrared (NIR) spectroscopy and multivariate calibration to predict stiffness and bulk density of 3.2 mm thick Yellow-Poplar veneer strips. Full-range (800-2500 nm) raw NIR spectra and spectra pre-treated using the first derivative method, along with spectra from three other different wavelength windows of 1200-2400 nm, 1800-2400 nm and 1400-2000 nm were regressed against the dynamic modulus of elasticity (stiffness; GPa) and the bulk density (kg/m 3) values of the veneers using the projection to latent structures (PLS) method to develop calibration models. All predictive models developed performed well in the prediction of stiffness and bulk density of new test samples that were not included in the calibration models. R 2 values ranged from 0.56-0.72 and 0.67-0.78 respectively for stiffness and bulk density. There was significant improvement in models developed with first derivative spectra over models developed with raw spectra. The models developed using the first derivative used fewer latent variables to achieve predictive models with higher R 2 values, lower root mean square errors of prediction (RMSEP) and standard errors of prediction (SEP). Models developed using the full NIR spectra range (800-2500 nm) and the NIR spectra region of 1200-2400 nm performed better than models developed using the restricted NIR wavelength regions of 1800-2400 nm and 1400-2000 nm. However, there was no clear distinction between models developed using the full NIR spectra range and the NIR spectra region of 1200-2400 nm. Overall, models developed with the first derivative pre-processed spectra using the whole NIR spectra performed best in predictability. The results of this study show the potential of using multivariate data analysis coupled NIR spectroscopy for on-line sorting and assessment of veneer stiffness prior to the lay-up process in the manufacturing of veneer-based engineered wood products such as plywood, Parallam and laminated veneer lumber.

The moisture content (MC) of wood affects its industrial performance, but it is difficult to monitor spatial variations in MC. Here, a multivariate regression was developed to estimate the MC from near infrared (NIR) spectra and was used to monitor the spatial variation in the MC of wood during air- and oven-drying. The spectra and mass of wood pieces were measured at five stages during drying (at each 20% loss of initial water mass). Wood pieces were dried naturally and oven-dried at 60 °C. Initially, 25 spectra were recorded at equidistant points covering the entire longitudinal × radial surface of the sample. Then, a planing machine was used to access the inside of the wood, and NIR spectra were measured for each new surface, at a total of 100 points spatially distributed within the wood pieces. The wood pieces were analyzed in their original state, and when they had lost 20, 40, 60, and 80% of their initial water mass. An NIR-based regression (R 2 p = 0.90 and RMSEP = 10.51%) was applied to estimate the MC, and its spatial gradient during drying was then mapped. These analyses revealed the spatial variation in MC within wood pieces during drying.