Exploring the potential of multispectral imaging for wood species discrimination

The recent publication of the European Regulation on deforestation, linked to the import and export from the European Union (EU) of certain commodities and products associated with deforestation and forest degradation, controls sustainable and legal sourcing of wood and wood-based products in EU. In this context, the accurate verification of wood species and their origin has become increasingly crucial. In this work, a multispectral camera was adopted to retrieve high resolution remotely sensed imagery of different wood samples exploring the spectrum between 440 and 860 nm. Eighteen wood species spectra were investigated. Starting from these spectra, linear discriminant analysis (LDA) proved that the band at 665 nm is the first discriminative feature, followed by 490 nm and 560 nm, respectively. Bands at 783 nm or higher wavelengths, i.e. the NIR region, discriminate selected species poorly. Using the first 4 linear discriminants, a classification of wood species was performed using the minimum Mahalanobis distance algorithm. The majority of species showed class accuracies between 0.7 and 0.9. However, some species showed poor performances. Cluster analysis involving all available spectra proved that the higher classification errors occurred between species of the same spectral cluster. This work shows the potentialities of adopting cheap and rapid screening tool (cameras) for separating selected wood species opening new scenarios to support industrial and commercial control processes.