论文部分内容阅读
提出了一种基于光谱纯度值的变量选择方法。对光谱中各变量计算其纯度值后,按降序将相应变量排列,采用PLS交互检验按前向选择法选择最佳变量子集。用本方法对3组分混合物体系及烟草样品的近红外光谱进行变量选择,并与全谱变量建模的结果进行了比较。结果表明本实验给出的波长变量优选方法是一种比较有效和实用的变量筛选方法,通过变量筛选,可极大地减少光谱信息重叠,从而提高定量校正模型的预测精度和建模效率。
A method of variable selection based on spectral purity is proposed. After calculating the purity value of each variable in the spectrum, the corresponding variables are arranged in descending order, and the optimal variable subset is selected according to the forward selection method by the PLS interaction test. The method was used to select the near-infrared spectra of the three-component mixture system and tobacco samples and compared with the results of the full-spectrum variable modeling. The results show that this method is a more effective and practical method for variable selection. Through variable screening, spectral information overlap can be greatly reduced, and the prediction accuracy and modeling efficiency of quantitative correction model can be improved.