Hyperspectral Imaging Predicts the Moisture Content Distribution in Acetylated Wood

The determination of moisture content across wood surfaces is important in many applications of wood products, but traditional (i.e. gravimetric) methods fail to detect moisture content variation in spatial dimensions. We reported the application of near-infrared (NIR) hyperspectral imaging to predict the moisture distribution on the wood surface at equilibrium. For this purpose, an experimental design was developed to create a large range of wood moisture contents by the variation of the acetylation degree of the woodblocks (0-17% WPG) and the relative humidity (0-95%) during their conditioning. The surfaces of the conditioned blocks were scanned with a hyperspectral NIR camera and a partial least square regression model was developed to predict the moisture content on the radial surface of the woodblocks. An external test image set was used to correctly identify the number of latent variables and the model parameters such as root mean square pixel error of a test image set was calculated. The results showed that hyperspectral NIR image regression can accurately predict variations in moisture content across the wood surface. In addition to sample-to-sample variation in moisture content, our results also revealed differences in the moisture content between early- and latewood in acetylated samples. This was in line with recent studies (Awais et al. 2020; Mäkelä et al. 2021), which found that thin-walled earlywood cells are acetylated faster. Our results demonstrated that the NIR hyperspectral imaging is an important tool to reveal the surface chemical information and can be adapted with certain upscaling in modern wood industries.