In this study, a method suitable for online rapid classification and separation of red oak and white oak wood species has been developed with the use of near infrared (NIR) spectroscopy and soft independent modeling of class analogies (SIMCA). The spectra of 150 wood specimens of each species were collected over wavelength window of 800–2500 nm. The raw spectra and spectra pre- treated by first derivative and standard normal variate (SNV) transformation were used to develop calibration models using the wavelength ranges 800–2500 nm, 1100–2200 nm and 1400–1900 nm by the SIMCA method. Principal component analysis (PCA) models were made for each class from the calibration set consisting of 100 specimens of red oak and white oak, respectively, and specimens not present in the calibration set (Testing set) consisting of 100 specimens were tested for classification according to the SIMCA method at the 5% and 25% significance level. Type I error (rejection of true member) associated with the models developed ranged from 2% to 20% and from 16% to 66%, respectively, for 5% and 25% significance level, while type II error (acceptance of a false member) was 0% for all models developed. There was no significant improvement in models developed with spectra pre-treated with first derivative or standard normal variate transformation over models developed with raw spectra. The full NIR spectral region of 800–2500 nm provided the most useful information for distinguishing between the two species. The results of this study showed that NIR spectroscopy coupled with the SIMCA method of multivariate data analysis could be used to reliably separate and sort wood of red oak and white oak.

Keywords: red oak, white oak, classification, near infrared (NIR), calibration set, testing set, principal component analysis (PCA) and soft ­independent modelling of class analogies (SIMCA)