期刊
CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS
卷 92, 期 2, 页码 125-130出版社
ELSEVIER SCIENCE BV
DOI: 10.1016/j.chemolab.2008.02.002
关键词
NIR spectroscopy; multivariate calibration; pre-processing methods; Scots pine; wood modification; furfurylation; weight percent gain; uncertainty estimation
Near-infrared spectroscopy was used to determine the degree of treatment in Scots pine solid wood modified with in situ cured furfuryl alcohol. The treatment range was approximately 0 to 30% weight gain relative to wood dry mass. Reflectance near-infrared spectra were recorded - using a conventional, large sample cup - directly on wood surfaces without sample preparation. Partial Least Squares regressions were used to construct a model for weight percent gain; initially for one data set but later extended to an external test set not treated fully in the same way as the original set. This gave rise to an increase in the prediction error which was sought neutralised by addition of a small fraction of the new samples in the original calibration set. Four spectral ranges and 13 pre-processing methods were tested and the final model with addition of 30% of the new samples in the original calibration set resulted in a model with a Root Mean Square Error of Prediction of 1.7 +/- 0.1 % weight gain and a R-2 of 0.97 +/- 0.01. (C) 2008 Elsevier B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据