制浆材木质素含量近红外分析模型传递研究

Study on Near-infrared Calibration Model Transfer for Lignin Content in Pulpwood

为研究制浆材中木质素含量近红外分析模型在两台便捷式近红外光谱仪间的传递,对制浆材木质素样品近红外光谱数据集进行代表性样本的选取、光谱预处理和界外样本的剔除,建立了源机的优化偏最小二乘(PLS)校正模型。分别采用斜率截距算法(S/B)、直接校正算法(DS)和典型相关分析算法(CCA)进行源机与目标机间的模型传递并比较了预测效果。结果表明,S/B算法模型传递效果较差,而经DS算法和CCA算法模型传递后的预测效果均有大幅提升。DS算法模型传递后决定系数(R^2)、预测标准差(RMSEP)和相对标准差(RPD)分别为0.9643、1.0370%和5.3513;CCA算法模型传递后R^2为0.9540、RMSEP为1.1766%、RPD为4.7711。因此,DS算法和CCA算法均可实现制浆材木质素含量近红外分析模型在两台便携式近红外光谱仪之间的传递。

The near-infrared calibration model transfer for lignin content in pulpwood was investigated between two portable near-infrared spectrometers.An optimal calibration model of master was established by partial least square(PLS)after the selection of representative infrared spectroscopy data net samples,preprocessing and eliminating outlier samples.The near-infrared spectroscopy calibration model was transferred between master and slave by the algorithms of slope/bias(S/B),direct standardization(DS)and canonical correlation analysis(CCA),respectively,and the prediction results were compared.The results indicated that the models transferred by DS and CCA improved the prediction accuracy significantly comparing to the algorithm of S/B with a poor performance.The coefficient of determination(R^2),root mean square error of prediction(RMSEP)and ratio of performance to standard deviate(RPD)by DS were 0.9643,1.0370%,5.3513,and by CCA were 0.9540,1.1766%,4.7711,respectively.Therefore,both DS and CCA algorithms could achieve the calibration model transfer between the two portable near-infrared spectrometers.