A system for optimal edging and trimming of rough hardwood lumber

Despite the importance of improving lumber processing early in manufacturing, scanning of unplaned, green hardwood lumber has received relatively little attention in the research community. This has been due in part to the difficulty of clearly imaging fresh-cut boards whose fibrous surfaces mask many wood features. This paper describes a prototype system that scans rough, green lumber and automatically provides an optimal edging and trimming solution along with resulting lumber grades. The system obtains thickness (profile) and reflectance information at 1/16-inch (1.6mm) resolution, using commercially available laser sources and a video camera. It analyzes the resulting images to detect wane and important lumber-degrading defects. A hierarchical defect detection scheme first analyzes the profile image for shape-based characteristics to locate wane, and to identify holes, splits, and background. Wane boundaries are detected with 3/16-inch (5-mm) error on average. The reflectance image is then assessed using a modular artificial neural network (MANN) to locate clear wood, knots, and decay. The MANN consists of a multilayer perceptron network for the detection of clear wood, and a statistically trained radial basis function network that identifies knots and decay. With this approach, we have achieved a pixel-level classification accuracy of 96.7%. Finally, a postprocessing step refines MANN output by identifying manufacturing marks that have been incorrectly classified as defects. Application software then finds optimal solutions for placement of cuts to yield maximum commercial value.