CT Image Analysis for the Identification and Three-Dimensional Reconstruction of Internal Log Characteristics of Sugar Maple (Acer saccharum) and Black Spruce (Picea mariana)

Internal log quality information helps optimize wood processing to improve product value from the logs. Computer tomography (CT) is a nondestructive technique that has been applied to produce internal log images. To date there has still been a lack of reliable image processing methods to accurately extract internal log information from log CT images. This research investigated the feasibility of identifying internal log characteristics in CT images using two promising methods, namely maximum likelihood classifier (MLC) and the back propagation artificial neural network (BP-ANN) classifier. Moreover, the marching cubes algorithm was adapted to reconstruct three-dimensional (3D) internal log images. Four log characteristics, including heartwood, sapwood, bark and knot in sugar maple and black spruce were considered in this study. The diagnostic value of major image features, including spectral, distance and textural features, in distinguishing the log characteristics were assessed. Useful image features were selected to develop the two types of classifiers. Hidden node number of the BP-ANN classifier was chosen based on the performance indicators: overall accuracy, mean square error, training iteration number, and training time. This research suggests that both MLC and BP-ANN classifiers achieved good classification performance on the two species. The BP-ANN classifier had better classification performance comparing with the MLC classifier. The sapwood of sugar maple and heartwood of black spruce logs used in this study are easier to identify. The results reveal that the separability of one log characteristic from the other log characteristics in CT images is mainly related to its physical properties including wood density and moisture content. The preliminary study on the 3D reconstruction of internal logs shows that the adapted method produced rough 3D log images. Accuracy analysis for the 3D reconstruction remains to be addressed in the future work.