基于近红外光谱技术的人工林杨木苯醇抽提物质量分数的预测

Prediction of Benzene-ethanol Extractive Content of Poplar Using Near Infrared Spectroscopy

用近红外光谱法对人工林杨木的苯醇抽提物质量分数进行了快速测定。用国家标准方法测定了42个杨木木材样品苯醇抽提物质量分数,并用近红外光谱仪(LabSpec Pro FR/A114260)测定相应的光谱。在350～2 500 nm、1 300～2 050 nm、2 050～2 500 nm 3个不同的光谱区域,采用未处理、Baseline、一阶导数、二阶导数等光谱的预处理方法,并采用PLS1、PLS2、PCR等3种不同的建模方法,建立了相应的校正模型与交互验证模型。结果表明:当光谱区域为2 050～2 500 nm、光谱数据进行一阶导数预处理、采用PLS1的建模方法、主成分数为7时,建立的校正模型预测效果最佳,校正模型的相关系数r=0.997 8、均方根误差为0.000 3、标准误差为0.000 3,验证模型r=0.736 0,均方根误差为0.006 1、标准误差为0.006 1;采用建立的模型对未参与建模的样本进行预测,其预测结果与实测结果之间的相关系数为0.918 6。

The benzene-ethanol extractive contents of 44 samples of poplar were determined by national standard of China, and then the near infrared （NIR） of all samples were collected by LabSpec Pro FR/Al14260. The calibration and validation model were built using PLS1, PLS2 and PCR with different pretreatment methods of un-pretreatment, baseline, the first derivative and the second derivative in different spectral region of 350-2 500 nm, 1 300-2 050 nm and 2 050-2 500 nm. The best model was built using PLS1 with the first derivative pretreatment of spectral data and seven principal components in 2 050-2 500 nm. The coefficients of correlation （r）, the root mean square error and the standard error of calibration model are 0. 997 8, 0. 000 3 and 0. 000 3, respectively, and 0. 736 0, 0. 006 1 and 0. 006 1 for validation model, respectively. The correlation （r） is 0. 918 6 between the prediction and lab measuring values.