Log Defect Recognition Using CT Images and Neural Net Classifiers

Although several approaches have been introduced to automatically identify internal log defects using computed tomography (CT) imagery, most of these have been feasibility efforts and consequent y have had several limitations: (1) reports of classification accuracy are largely subjective, not statistical, (2) there has been no attempt to achieve real-time operation, and (3) texture information has not been used for segmentation, but has been limited to recognition procedures. Neural network classifiers based on local neighborhoods have the potential to greatly increase computational speed, can be implemented to incorporate textural features during segmentation, and can provide an objective assessment of classification performance. This paper describes a method in which a multilayer feed-forward network is used to perform pixel-bypixel defect classification. After initial thresholding to separate wood from background and internal voids, the classifier labels each pixel of a CT slice using histogram-normalized values of pixels in a 3×3×3 window about the classified pixel. A postprocessing step then removes some spurious pixel misclassifications. Our approach is able to identify bark, knots, decay, splits, and clear wood on several species of hardwoods. By using normalized pixel values as inputs to the classifier, the neural network is able to formulate and apply aggregate features, such as average and standard deviation, as well as texture-related features. With appropriate hardware. the method can operate in real time. Sawmill operators must confront a number of drastic changes to traditional ways of operating their mills. These changes have been precipitated by expanded markets (both export and domestic), low-quality raw material, increased competition from nonwood products, social pressures to manage public lands for nontimber resources, and reduced profit margin between log costs and lumber prices (Schmoldt 1993). Consequently, sawmills need to improve their operations in several ways. They must consume less raw material while producing an equivalent amount of final product, and manufacturing output must be more consistent and of higher value. This implies that mill operations must scrutinize carefully the breakdown of logs into lumber, with the intent to make that conversion as efficient as possible. Knowledge of internal log defects, obtained by scanning, is a critical component of any such efficiency improvements (Occeña 1991). Before computed tomography (CT) scanning or any other type of internal log scanning can be applied in industrial operations, there are several hurdles that must be overcome. First, there needs to be some way to automatically interpret scan information so that it can provide the saw operator with information needed to make proper sawing decisions. A sequence of x-ray tomographs cannot be readily synthesized into a three-dimensional (3D) mental model by human operators (Schmoldt and others 1993). For the purposes of sawing the log cylinder into high-value boards, this means accurately locating, sizing, and labeling internal defects. Second, this defect recognition procedure must operate at real time speeds, so that scanning, image reconstruction, and image interpretation and display can be integrated into mill processing. Third, a 3D display of a log and its defects for the sawyer is only the first step toward real