最小相关系数的多元校正波长选择算法

在近红外光谱的定量分析中,由于仪器的精密程度越来越高,采集的光谱数据通常具有很高的维度.因此,波长选择对于剔除噪声及冗余变量,简化模型,提高模型的预测性能是必不可少的.近红外光谱特征波长选择方法众多,但变量间的多重共线性问题仍是导致模型效果较差的一个关键问题.变量间共线性可以通过相关系数进行分析,当相关系数高于0.8,...

Wavelength Selection Algorithm Based on Minimum Correlation Coefficient for Multivariate Calibration

In the quantitative analysis of near-infrared spectroscopy, as the instrument’s precision is getting higher and higher. The collected spectral data usually has a very high dimension. Therefore, wavelength selection is essential for eliminating noise and redundant variables, simplifying the model, and improving the model’s predictive performance. There are many methods for selecting characteristic wavelengths in NIR spectroscopy, but the problem of multicollinearity among variables is still a key issue that leads to poor model effects. Collinearity between variables can be analyzed by correlation coefficient. When the correlation coefficient is higher than 0.8, it indicates that there is multicollinearity. Therefore, this paper takes the correlation coefficient between variables as the selection criteria and proposes a wavelength selection method that minimizes the collinearity between the selected variables, called the Minimal Correlation Coefficient (MCC) method. This method is based on the correlation coefficient matrix of the spectrum data. It selects the wavelength with the smaller average and standard deviation of the correlation coefficients of other wavelengths as the candidate modeling wavelength set so that the linear correlation between the wavelengths in the set is minimized, and the model has eliminated Collinearity between variables. Then use the standard regression coefficient to select the wavelength that has a greater impact on the dependent variable to obtain the prediction model. In order to verify the effectiveness of the proposed algorithm the method is tested. Using two sets of opening NIRS data sets (diesel dataset and soil dataset), wavelength selection was carried out by MCC algorithm, and compared with several other commonly used wavelength selection methods, including successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), random frog (RF) and iteratively retains informative variables (IRIV). The experimental results show that the MCC algorithm has good prediction performance, the model prediction accuracy based on MCC is better than that of SPA, CARS, RF, and is roughly the same as that of IRIV. Therefore, the minimum correlation coefficient method is an effective wavelength selection algorithm, which can reduce the dimension efficiently and improve the prediction precision of the model.