小波变换结合多维偏最小二乘方法用于近红外光谱定量分析

Quantitative Analysis of Near Infrared Spectroscopy by Combination of Wavelet Analysis and N-way Partial Least Square

将小波变换和多维偏最小二乘法相结合用于近红外光谱定量校正模型的建立。首先将原始光谱进行小波变换分解,得到系列小波细节系数,通过选取一组受外界因素少、信息强的小波系数组成三维光谱阵,然后再采用多维偏最小二乘法建立校正模型。实验结果表明,该方法所建近红外校正模型的预测能力更强,并更具稳健性。

A new multivariate calibration method ior near infrared spectroscopy quantitative analysis was proposed based on wavelet analysis and N-way partial least square method. After the decomposition of the raw spectrum by wavelet analysis, a set of featured wavelet detail coefficients were selected and reconstructed to yield a three-way matrix. The N-way partial least square method was used to develop calibration model on the basis of the three-way matrix. Significant improvement of predictive capability over results from traditional partial least square method and wavelet analysis has been observed in the gasoline data set for prediction of research octane number. In addition, the calibration model obtained by the proposed method is more robust against unexpected spectral variations such as caused by sample temperature.