Potential of machine learning approaches for predicting mechanical properties of spruce wood in the transverse direction

Abstract To predict the mechanical properties of wood in transverse direction, this study used machine learning to extract anatomical features from cross-sectional stereograms. Specimens with different orientations ray parenchyma cell were prepared, and their modulus elasticity (MOE) rupture (MOR) measured by a three-point bending test. The orientation density ( $$\rho $$ ρ ) as parameters for MOE MOR prediction. Conventional algorithms artificial neural network used, satisfactory results obtained both cases. A regular convolutional (CNN) density-informed CNN automatically specimens’ stereograms properties. achieved acceptable but relatively low accuracy reason may be that information could not satisfactorily extracted images, because images represented limited region specimen. For CNN, average prediction coefficient drastically increased when was provided. regression activation map constructed understand representative are strongly related latewood highly activated It is believed ratio successfully considered different. earlywood prediction, while transition between These provide new insights into relationship wood.