一种改进小波算法及其在拉曼光谱预处理的应用

Improved Wavelet Transform Algorithm and Its Application in Raman Spectra Preprocessing

针对现有用于光谱预处理的小波变换算法对光谱噪声和背景荧光等处理效果不佳的局限性,本文提出了一种改进的小波变换算法——小波变换频率分量相关选择法,首先对拉曼光谱进行小波棱镜分解,然后计算各个频率分量与待测质量指标的相关系数,设定相关系数的相对阈值,提取高于阈值的小波频率分量波长点光谱数据作为校正模型的有效输入数据。将其应用于汽油低分辨率拉曼光谱的预处理,并采用预处理后光谱建立的偏最小二乘模型预测值的最大正负误差和交叉检验的均方误差作为指标。实验结果证明,与其他常见预处理方法比较,该方法并能够很好地减弱荧光背景干扰和高频噪声,显著提高了基于偏最小二乘方法建立的汽油辛烷值的模型预测精度,其均方误差减少为0.23;此外,采用该预处理方法的偏最小二乘模型的均方误差随主元数变化不大,稳健性也比采用其他预处理方法的效果好。

To overcome the limitations of existing wavelet transform （WT） preproeessing methods for Raman spectra, such as bad performance on spectral noise and fluorescence, an improved preprocessing method-WT frequency component correlative selection algorithm was proposed. In this method,Raman spectra are firstly prism-decomposed by WT,then correlations between every frequent weight and target are computed and threshold is set to select the efficient input data for calibration model. This method is applied in gasoline Low-Resolution Raman spectra data preprocessing; the max positive/negative error and root mean squares error of cross validation （RMSECV） of the partial least square （PLS） model based on spectra after preprocessing is used to build are selected as criterion. Compared with other existing method, the experimental results show the new algorithm obviously weakens the fluorescence and high frequent noise and improves the prediction performance of the PLS model for gasoline octane number, the RMSECV can be reduce to 0. 23; besides, the RMSECV of PLS model based on proposed method does not change dramatically along with the change of the latent number of PLS model. So this method is more robust than others.