苹果可溶性固形物含量的近红外模型传递研究

Study on near-infrared calibration model transfer for soluble solid content in apple

研究了苹果可溶性固形物含量(SSC)的近红外模型传递问题。样品集经过异常剔除和预处理之后,建立了主机的偏最小二乘(PLS)预测模型。分别采用一元线性回归直接标准化(SLRDS)、斜率/截距(S/B)法和专利算法(Shenk's)进行主机和从机间的模型传递,讨论了3种算法的传递效果和转换集标样数对传递效果的影响。结果表明,经过3种方法传递后预测集的预测效果比传递前均有不同程度提高,其中SLRDS算法传递效果最佳,经SLRDS传递后相关系数(R)、预测标准偏差(RMSEP)、相对分析误差(RPD)和光谱平均差异(ADS)分别为0.9614,0.4735,3.3866和0.0661。因此利用SLRDS算法实现两台便携式光谱仪之间的模型传递是可行的。

The problem apples was investigated of near-infrared calibration model transfer for soluble solid content （SSC） of in this paper. A partial least squares regression （PLS） model was established on the master after eliminating outlier samples and preprocessing. The established calibration model was transferred by simple linear regression direct standardization （SLRDS） algorithm, slope/bias （S/B） algorithm and Shenk＇ s algorithm, respectively. The effects of the number of standard samples and transferring efficiency were discussed. The results indicated that the predictive results had been improved by three algorithms compared to the results before transfering. But prediction accuracy by SLRDS algorithm is higher than the other two methods, and correlation coefficient （ R）, root mean squared errors of prediction （RMSEP）, relative prediction deviation （RPD） and average diversity of spectra （ADS） by SLRDS were 0. 9614, 0. 4735, 3. 3866 and 0. 0661, respectively. Thus the SLRDS algorithm can be used to successfully transfer the calibration model between two near-infrared instruments.