Data analysis of hyperspectral NIR image mosaics for the quantification of linseed oil impregnation in Scots pine wood

Abstract

Eight pieces of sapwood and heartwood from Scots pine, before and after linseed oil impregnation, were used to make 16 near-infrared hyperspectral images (90–200 pixels wide × 466–985 pixels long × 239 wavelengths: 982–2,480 nm). The wood pieces were selected according to a 2 × 2 experimental design using radial–tangential cut and heartwood–sapwood as factors with two replicates. A first mosaic of 16 images was cleaned and analyzed by image principal component analysis. Interpretation was realized by studying score images and score plots by brushing interaction. In the resulting T1–T2 score plot, the untreated pieces formed a dense cluster, while the impregnated ones showed larger variation. The good separation of treated and impregnated clusters was confirmed by PLSDA showing low false negatives and positives. Analysis of the eight impregnated wood pieces clearly showed regions of wrong impregnation in one wood piece. Loadings resembling linseed oil spectra indicated that this was due to badly polymerized linseed oil. After removing the outlier piece, a new model was made on the seven-piece mosaic showing in the T1–T2 score plot that heartwood and sapwood absorbed the linseed oil differently. This difference was not detected in the untreated wood, so it had to come from the impregnation process. Edges reacted differently from surfaces to the impregnation process as seen in the T1–T4 score plot. These findings show that a future online quality inspection of both raw wood and impregnated pieces would be feasible.