Technical framework of advanced volumetric sub-μ-CT wood imaging integrated with adaptable deep-learning-based wood species classification: initial evaluation on softwood and hardwood data

Summary In an age of increasing surveillance and control of markets and trade flows, new methods for species identification of commercial timbers and other woody plant products are needed. Traditional plant anatomy has generated a vast body of knowledge on taxonomical identification. In highly processed and mixed materials such as charcoal, fibreboard board and paper, where DNA and chemotaxonomic composition is regularly destroyed or altered, the anatomical structure is often still intact. Comprehensive studies in the field of state-of-the-art μCT analysis have shown that the use of sub-micrometre CT scanning technology can be very suitable for wood identification and other structural investigations. The current state of CT technology (resolution, working speed, feasibility) opens up various options for use in combination with other relevant key technologies, in particular machine learning and other forms of digitisation in the context of wood research. Thus, the primary objective of this methodologically oriented study is to design and implement, for the first time, an automated and modular classification pipeline that integrates machine learning with volumetric high-resolution sub-μCT imaging for wood identification. To evaluate this approach, we conducted initial experiments using a simplified basal case, applying the method to sub-μCT scans of softwood ( Pinus sylvestris ) and hardwood ( Acer pseudoplatanus ).