鱼粉近红外光谱模型传递应用研究

Spectral Transfer of Near-infrared Spectrometric Model for Fish Meal

为探讨光栅型与傅里叶变换型近红外分析仪之间模型传递的应用效果,选取国产鱼粉为近红外光谱样本,DS2500F型近红外分析仪为源仪器,MPA型近红外分析仪为目标仪器,采用分段直接校正(PDS)方法实现近红外光谱传递。分别建立水分、粗蛋白质、粗脂肪、蛋氨酸和赖氨酸等组分的预测模型,通过交互验证决定系数(Rcv^2)、交互验证标准误差(RMSECV)、马氏距离(MD)、系统偏差(Bias)、预测均方根误差(RMSEP)和相对分析误差(RPD)等参数,多维度评估光谱传递后所建预测模型的效果。结果表明,DS2500F仪器的近红外光谱传递到MPA型仪器时,所建国产鱼粉的水分、粗蛋白质、粗脂肪、蛋氨酸、赖氨酸的预测模型与MPA型仪器原始预测模型各参数对比无显著差异,预测效果基本一致,说明国产鱼粉在DS2500F仪器上的近红外光谱通过传递可以替代MPA型仪器的原始光谱,间接实现了模型传递,且具有良好的适用性和共享性,可提高近红外预测模型的应用效率。

In order to investigate the feasibility of spectral transfer between near-infrared spectrometric(NIRS) models of grating type and Fourier transform near infrared analyzer,the spectral transfer of original spectrum was performed by piecewise direct standardization(PDS) method,with domestic fish meal as the NIRS sample,DS2500 F as the source instrument and MPA as the target one.The NIRS quantitative analysis prediction models for moisture,crude protein,crude fat,methionine and lysine were established,whose spectrum difference,applicability and accuracy were verified by correlation coefficient of cross validation(Rcv^2),root mean square error of cross validation(RMSECV),Mahalanobis distance(MD),bias,root mean square error of prediction(RMSEP) and relative prediction deviation(RPD),respectively.According to the verification and evaluation results,the various parameters of the prediction models established for moisture,crude protein,crude fat,methionine and lysine were not significantly different from the NIR spectrum of DS2500 F instrument transferring to the target spectrum of MPA instrument.It showed that the NIR spectrum of domestic fish meal on the DS2500 F instrument could be used to replace the original MPA spectrum by transmission,and model transfer was thus realized indirectly,which improved the application efficiency of NIR prediction model with good shareability and applicability.