Convolutional Neural Networks for Knot Measurement in Tomographic Images of Wood Logs

This work provides novel approaches to knot diameter and DKB coordinate regression in wood logs by exploiting Convolutional Neural Networks (CNNs). One of the proposed procedures, concerning the subsequent application of a CNN for status classification to CT voxel sub-blocks containing consecutive portions of each knot, yields promising results and is computationally fast enough to be employed in in-line applications, both in terms of diameter regression and status estimation.