论文部分内容阅读
为了实现便携式近红外光谱仪现场快速分析小麦粉中灰分的含量,对125个小麦粉样本扫描并进行多种预处理后,建立了基于偏最小二乘(PLS)的定量分析模型。探讨了基线校正(Baseline)、矢量归一化(Normalize)、SavitskyGolay卷积平滑法、导数、标准正态变量变换(Standard Normal Variate Correction,SNV)以及多元散射校正(Multiplicative Scatter Correction,MSC)这六种预处理方法及其组合方式对建模的影响。结果表明:矢量归一化+Savitsky-Golay滤波平滑法是最佳预处理方法,相应建立的小麦粉灰分含量最佳模型的校正决定系数R_c~2为0.947,交叉验证决定系数R~2v为0.896,校正均方根误差(RMSEC)为0.026,交叉验证均方根误差(RMSECV)为0.037,预测均方根误差(RMSEP)为0.026。无预处理模型的校正决定系数为0.873,交叉验证决定系数为0.832,校正均方根误差为0.044,交叉验证均方根误差0.051,预测均方根误差为0.056;相较于无预处理模型,最佳模型的预测精度和稳健性有了很大的提高。
In order to quickly analyze the content of ash in wheat flour by portable near infrared spectrometer, a quantitative analysis model based on partial least squares (PLS) was established after scanning and processing 125 wheat flour samples. Baseline, Normalize, SavitskyGolay convolution smoothing, derivatives, Standard Normal Variate Correction (SNV) and Multiplicative Scatter Correction (MSC) Effects of Pretreatment Methods and Their Combination Methods on Modeling. The results show that the vector normalization + Savitsky-Golay filtering smoothing method is the best pretreatment method, and the best determination coefficient R_c ~ 2 is 0.947, the cross validation coefficient R ~ 2v is 0.896, The root mean square error of correction (RMSEC) was 0.026, the root mean square error of cross validation (RMSECV) was 0.037, and the root mean square error of prediction (RMSEP) was 0.026. The calibration coefficient of the non-pretreatment model was 0.873, the coefficient of cross validation was 0.832, the root mean square error of calibration was 0.044, the root mean square error of cross validation was 0.051, and the root mean square error of prediction was 0.056. Compared with no pretreatment model, The prediction accuracy and robustness of the best model have been greatly improved.