基于拉曼光谱和多空间融合预处理方法的葡萄糖成分检测

Multi-Scale Pre-Processing for Raman Quantitative Analysis of Glucose

采用多空间融合的预处理方法结合偏最小二乘回归,对低浓度葡萄糖样品的拉曼光谱进行定量分析。通过傅里叶变换拉曼光谱仪获得低浓度葡萄糖样品的光谱数据,结合多空间融合预处理方法和采用传统的偏最小二乘方法建立建立线性定量回归分析模型。用相关系数r、预测均方根误差RMSEP、交叉验证均方根误差RMSECV等参数作为评价模型的指标,实验表明,不同的光谱预处理方法对建模效果有较强的的影响,通过比较可知将基线校正、正交信号校正及Savitsky-Golay平滑三种预处理结合后的,模型的r=0.979 8,RMSEP=0.031 7,RMSECV=0.031 0,明显优于其他预处理方式。由此说明,通过多空间融合的预处理方法对光谱数据进行优化的操作是可行的,较好的预处理融合方式对实验模型准确性和稳健性影响很大。

In this paper,we used Raman spectroscopy system to measure the spectral data of low concentration glucose sample solution.After the original data under the different preprocessing methods and their recombination,we built the same regression model method called partial least-squares regression method to compare the values of some parameters,such as coefficient of determination(r),cross validation root mean square error(RMSECV),the root mean square error of prediction(RMSEP),to get the more prominent optimization condition.The experiment results show that the preprocessing using Baseline Correction,Orthogonal Signal Correction and Savitsky-Golaysmooth can get the data indicators with r=0.979 8,RMSECV=0.031 7,RMSEP=0.031 0,which is better than other preprocessing methods,so as to reduce the interference without loss of sample characteristic information.Therefore,it is feasible to optimize the spectral data through a multi-space fusion pre-processing method,and a good pre-processing fusion method has a great influence on the accuracy and robustness of the experimental model.