Classification of wood surface features by spectral reflectance

A database of spectral-reflectance curves of Douglas-fir veneer surface features is presented and analyzed via principal-component analysis. The paper describes how such analysis can be used to model and classify the spectral-reflectance curves by feature type. For modeling (curve-reconstruction) purposes, three principal components were sufficient by most criteria. For classification purposes, seven principal components achieved classification accuracies (with quadratic discriminant analysis) on the order of 99%, comparable to the accuracies achieved with the raw spectral data. The best seven principal components were not those associated with the largest variation in the data. This paper suggests how comparable classification accuracies might be achieved in a system operating at production speeds in a mill.