CT Imaging of Hardwood Logs for Lumber Production
Daniel L. Schmoldt, Pei Xian Li, Amos Lynn Abbott · 1996
Hardwood sawmill operators need to improve the conversion of raw material (logs) into lumber. Internal log scanning provides detailed information that can aid log processors in improving lumber recovery. However, scanner data (i.e. tomographic images) need to be analyzed prior to presentation to saw operators. Automatic labeling of computer tomography (CT) images is feasible, but no research has established labeling accuracy or demonstrated real time operation. An automated labeling scheme is presented in this paper that is both very accurate and very fast. The procedure segments and classifies each pixel in a CT image as either knot, split, bark, decay, or clear wood by using a small 3D pixel neighborhood as input to an artificial neural network classifier. Initial results with two species of oak and with yellow poplar indicate that species-dependent classifiers of this type can be constructed. This classification approach can also be applied to other types of images encountered in industrial inspection applications, e.g., gray-scale and color images. KEYWORDS Classification, image processing, machine vision, industrial inspection, artificial neural network.