Prediction of wood properties for thawed and frozen logs of quaking aspen, balsam poplar, and black spruce from near-infrared hyperspectral images

This study tested the use of near-infrared hyperspectral images to estimate moisture content (MC) and basic specific gravity (BSG) of thawed and frozen logs of three species: quaking aspen, balsam poplar, and black spruce. For each species, more than 90 small 4 cm cubic samples were prepared and subjected to drying steps in both frozen and thawed conditions. At each step, hypercube images and sample weights were recorded to determine MC and BSG of each sample. Partial least squares (PLS) models were calibrated by considering two factors: log state (thawed and frozen conditions) and species, and their combination. With respect to the species, the PLS model accuracy depends on the range of variation in the input data. The model accuracy was the best for black spruce samples that have the lowest range of variation for both MC and BSG, whereas the model accuracy was the lowest for balsam poplar samples that have the highest range of variation for both MC and BSG. With all the data, the accuracy of the MC model worsened, but the accuracy of the BSG model reached a maximum (\( R_{\text{Validation}}^{2} \) = 0.88). The best PLS model was then employed to produce 2D MC and BSG images over the whole log disks. PLS discriminant analysis was also applied to sort the samples according to three MC or BSG classes, the species, and the log state (frozen and thawed). The overall accuracy was higher than 72 % for both the MC and BSG sorting, 86 % for the species sorting, and 97 % for the log state sorting.