鸡腿肌冻干粉蛋氨酸近红外光谱定量预测模型的建立与优化

Development and optimization of the near infrared spectroscopy quantitative prediction model for methionine of freeze-dried chicken leg muscle powder

中文关键词:  近红外光谱  鸡腿肌  冻干粉  蛋氨酸  定量预测模型

英文关键词:near infrared spectroscopy  chicken leg muscle  freeze-dried powder  methionine  quantitative prediction model

中文摘要:

      目的 建立和优化鸡腿肌冻干粉蛋氨酸近红外定量预测模型。方法 以263个鸡腿肌冻干粉近红外光谱(near infrared spectroscopy, NIRS)和蛋氨酸含量为研究对象, 分别采用7种不同光谱预处理方法、4种特征光谱筛选方法、2种蒙特卡洛交叉验证法(异常样本剔除方法)(monte carlo cross validation, MCCV)、样品集划分方法(sample set partitioning based on joint x-y distance, SPXY), 应用偏最二小乘法(partial least squares regression, PLS)、内部交互验证和外部验证建立和优化鸡腿肌冻干粉蛋氨酸近红外定量预测模型。结果 最优鸡腿肌冻干粉蛋氨酸NIRS定量预测模型在1000~2502 nm谱段, 使用原始光谱, 在标准正态变换结合一阶gapsegment (1#,15,7)导数法[standard normal variate-first order gapsegment derivative, SNV+gapsegment (1#,15,7)]光谱的基础上使用MCCV删除54个样本后, 采用基于x-y联合距离的SPXY选取156个校正样本, 39个外部验证样本所建模型, 其校正决定系数(coefficient of determination of calibration, R_CAL^2)为0.93, 内部交互验证决定系数(coefficient of determination of cross validation, SECV)为0.0609, 预测决定系数(coefficient of determination of prediction, R_P^2)为0.83, 验证集标准偏差与预测标准偏差的比值(ratio performance deviation calculated as SD/Sep, RPDP)为2.42。结论 模型预测值与化学检测值有很高的相关度, 对鸡腿肌冻干粉蛋氨酸NIRS模型预测精度和稳健性影响最大的因素是异常样本剔除方法和建模样本选取方法。

英文摘要:

      Objective To establish and optimize the near infrared quantitative prediction model for methionine of freeze-dried chicken leg muscle powder. Methods Taking the near infrared spectroscopy (NIRS) and methionine content of 263 chicken leg muscle lyophilized powder as the research object, 7 kinds of different spectral pretreatment methods, 4 kinds of characteristic spectral screening methods, 2 kinds of monte carlo cross validation (MCCV), the sample set partitioning based on joint x-y distance (SPXY), and partial least squares regression (PLS), internal interactive verification and external verification were used respectively to establish and optimize the near infrared quantitative prediction model for methionine of freeze-dried chicken leg muscle powder. Results The best model of the near infrared quantitative prediction model for methionine of freeze-dried chicken leg muscle powder was the original spectrum in the 1000-2502 nm spectrum. The 54 samples were deleted by using MCCV on the basis of standard normal transformation combined with the first-order gapsegment derivative (1#,15,7), 156 correction samples and 39 model built by external validation samples were selected by using SPXY method based on joint x-y distance. The coefficient of determination of calibration (R_CAL^2), coefficient of determination of cross validation (SECV), coefficient of determination of prediction (R_p^2), and the ratio performance deviation calculated as SD/Sep (RPDP) of the model was 0.93, 0.0609, 0.83, 2.42, respectively. Conclusion There is a relatively high degree of correlation between the predicted value by the NIRS quantitative prediction model and detection value by chemical method. The prediction accuracies and robustness of the model are mainly affected by the method of outlier samples elimination and the modeling sample selection method.