矩阵的奇异值分解 在红外光谱预处理中的应用

研究了矩阵的奇异值分解在红外光谱预处理中的应用。对奇异值序列,采用间距最大最高峰原则,选择有效秩的阶次;对原始光谱进行奇异值降噪处理、一阶微分处理,用相关系数法选择建模谱段为3 100~2 650 cm~(-1)和1 600~1 250 cm~(-1);采用PLS法建立运动粘度的校正模型,并与卷积平滑等预处理方法进行比较。模型校正集的相关系数(R_c)为0.977 9,标准偏差SEC为0.100 5,预测集的相关系数(R_P)为0.941 2,标准偏差SEP为0.154 7。研究结果表明:奇异值分解和一阶微分相结合可有效去除光谱噪声和基线漂移的干扰,提高PLS分析模型的准确度。

Application of SVD Denoising in the Preprocessing of Infrared Spectrum

The application of SVD denoising was studied in the preprocessing of infrared spectrum. The principle of largestspacing and maximum peak was adopted to choose effective rank order. The original spectrum was preprocessed by SVDdenoising and the first derivative, the wave wand were selected with the method of correlation coefficient. Calibration mod.el for kinematic viscosity was established using the PLS method, and was compared with convolution smoothing. Its Rc was 0.9779, its SEC was 0.1005, its Rp was 0.9412, its SEP was 0.1547. The results showed that SVD and the first derivativecould effectively conduct denoising, correct baseline drift and improve the accuracy of PLS model.