Knot Identification from CT Images of Young Pinus sylvestris Sawlogs Using Artificial Neural Networks

Abstract

The value of solid wood products is to a large extent determined by the sizes, types and distribution of the knots in the products. Hence there is a great interest in describing the internal knot structure of individual logs. The Swedish Stem Bank has been extensively used for modelling the interior knot structure of Scots pine ( Pinus sylvestris L.) and for simulating the outcome of sawing operations. The stem bank holds parametric descriptions, extracted from computer tomography (CT) imagery, of mature trees. To enlarge the stem bank with trees from younger stands, a better method for extracting the knot properties from the CT images is needed. In this study, artificial neural networks were used for segmenting and classifying knots in transverse CT images of a 30-year-old Scots pine. The cross-validated prediction rate of correctly classified pixels was 95.9% - 1.2%. Classified knots were distinctly separated. Misclassifications were mainly located in the border areas between knots and clear