Near Infrared Image Analysis for Online Identification and Separation of Wood Chips with Elevated Levels of Extractives

Abstract

Forest-based biorefinery feedstocks are usually broken up into wood chips prior to any form of processing. These wood chips have a complex and highly variable composition, although they may look identical to an inexperienced observer. Some chips have high contents of valuable extractives. Therefore, it would be desirable to separate such chips that are rich in extractives. Various fractions of pine and spruce wood were used to acquire near infrared (1000–2498 nm) hyperspectral images in order to explore the usefulness of multivariate image analysis for detection and separation purposes. Multivariate modelling by image principal component analysis detected large variations in extractive content among wood chips of different biomass types, for example, sapwood, heartwood and knotwood. The extractive parts could be classified in the images and their content could be reasonably well predicted. Partial least squares (PLS) regression models could be made between collected spectra and measured extractive contents. These worked better for milled and homogenised bulk samples than for average image spectra. Regression coefficients showed that the C–H bonds in the spectra were responsible for the validity of the models. The average image PLS models could be used to make prediction images showing the location of the regions with high extractive content in knotwood. The results indicate that extremely rapid spectral-based fractionation could be used to separate tailored biomass streams of wood chips.