Identification of selected internal wood characteristics in computed tomography images of black spruce: A comparison study

Abstract

The feasibility of identifying internal wood characteristics in computed tomography (CT) images of black spruce was investigated using two promising classifiers: The maximum likelihood classifier (MLC) and the back propagation (BP) artificial neural network (ANN) classifier. Nine image features including one spectral feature (gray level values), a distance feature, and seven textural features were employed to develop the classifiers. The selected internal wood characteristics to be identified included heartwood, sapwood, bark, and knots. Twenty cross-sectional CT images of a black spruce log were randomly selected to develop the two classifiers. The results suggest that both classifiers produced high classification accuracy. Compared with the MLC classifier (80.9% overall accuracy), the BP ANN classifier had better classification performance (97.6% overall accuracy). Moreover, statistical analysis reveals that the heartwood of the black spruce log used in this study is the easiest to identify by either classifier compared with the other three log features. The results also suggest that the separability of one wood characteristic from the other wood characteristics in black spruce CT images is mainly related to moisture content. © The Japan Wood Research Society 2009.