3步混合变量选择策略在鸡肉近红外水分检测中的应用

基于光谱多元校正中有效变量选择的3步混合策略(初筛、精挑、细选),提出了间隔偏最小二乘(iPLS)、区间变量迭代空间收缩法(iVISSA)和迭代保留信息变量(IRIV)联用的特征变量选择方法,对生鲜鸡胸肉的近红外光谱进行特征波长选择,建立了鸡肉水分检测模型。结果表明,建模波长数量经iPLS-iVISSA-IRIV 3步选择后减少为全光谱建模的0.76%,但模型精确度和稳定性逐步提高。选定8个特征波长建模,其校正相关系数RC=0.907 7,校正均方根误差RMSEC=0.516 1;预测相关系数RP=0.943 5,预测均方根误差RMSEP=0.612 3。表明基于3步混合策略提出的iPLS-iVISSA-IRIV方法能有效选择鸡肉水分检测的特征波长。

Research on the application of three-step hybrid variable selection strategy in chicken moisture detection by near infrared

Based on the three-step hybrid strategy of effective variable selection (rough selection, fine selection and optimal selection) in multivariate calibration of spectra, a feature variable selection method combining interval partial least squares (iPLS), interval variable iterative space shrinkage approach (iVISSA) and iteratively retaining informative variables (IRIV) was proposed to select the feature wavelengths of the near-infrared spectra of fresh chicken breast,and established a chicken moisture detection model. The number of modeled wavelengths was reduced by 0.76% after iPLS-iVISSA-IRIV, but the accuracy were stability of the model are gradually improved. The modeling results using the selected 8 characteristic wavelengths were as follows: correlation coefficient of calibration R_C=0.907 7, root mean square error of calibration RMSEC=0.516 1, correlation coefficient of prediction R_P=0.943 5, root mean square error of prediction RMSEP=0.612 3. The result shows that the iPLS-iVISSA-IRIV method based on the three-step hybrid strategy can effectively select the characteristic wavelengths of chicken moisture detection.