Rapid identification of wood species by near-infrared spatially resolved spectroscopy (NIR-SRS) based on hyperspectral imaging (HSI)

摘要： Conventional near-infrared (NIR) spectroscopy has shown its potential to separate wood species non-destructively based on the aggregate effect of light absorption and scattering values. However, wood has an aligned microstructure, and there is a large refractive index (RI) mismatch between the wood cell wall substance (n≈1.55) and the cell lumen (air≈1.0, water≈1.33). Light scattering is dominant over absorption μ′ (cid:31)( ) a in wood, and this fact can be utilized for complex classification purposes. In this study, an NIR hyperspectral imaging (HSI) camera combined with one focused halogen light source (? 1 mm) was designed to evaluate the light scattering patterns of five softwood (SW) and 10 hardwood (HW) species in the wavelength range from 1002 to 2130 nm. Several parameters were combined to improve the data quality, such as image histogram plots of defined spaced bins (associated with diffuse reflectance values of light), variance calculation on the frequency (the number of pixels in each bin) of each histogram and the principal component analysis (PCA) of all the variance values at each wavelength. The identification accuracy of the quadratic discriminant analysis (QDA) under the five-fold cross-validation method was 94.1%, based on the first three principal component (PC) scores.