Abstract:
Hardwood lumber is a major forest product, and board grading is an important part of its manufacturing and marketing. Computer grading programs have been used to train graders and to grade lumber for board data banks, but they have not been used to machine-grade boards in an industrial environment because of their slow performance. The Hardwood Lumber Grading Program (HLGP) was developed to quickly estimate a board's grade according to the National Hardwood Lumber Association rules by employing a pair-wise heuristic adopted from a separate board
cut-up program. This paper describes how HLGP, using a board's size and defect data, locates the board's non-conflicting, clear-wood regions, uses them to calculate the yield, compares the board's size and defects with the limitations in the grading
rules, and then assigns a grade. It also discusses how the original pair-wise algorithm was modified to expand the solution space without introducing too large a time complexity. Using a large board set that had been extensively evaluated both manually and by the most widely recognized computer grading program, HLGP
correctly graded 1,289 out of the 1,581 boards. This translates into a misclassification rate of 18 percent, which is too high for industrial applications. However, HLGP averaged only less than a second to grade a board creating ample opportunity to improve the algorithm's accuracy before increased execution time
becomes a problem.
