苹果产地差异对可溶性固形物近红外光谱检测模型影响的研究

Assessment of Influence of Origin Variability on Robustness of Near Infrared Models for Soluble Solid Content of Apples

为更好地利用近红外光谱预测苹果可溶性固形物含量,减少产地差异对近红外光谱检测模型的影响,以4种不同产地的富士苹果为研究对象,采用基于x-y共生距离的样本划分方法分别对不同产地的苹果选取代表性样本作为校正集,利用偏最小二乘算法,建立和比较单一产地和混合产地下的苹果可溶性固形物近红外光谱检测模型,并结合竞争性自适应重加权算法(CARS)和连续投影算法(SPA)对苹果可溶性固形物的建模变量进行筛选。相比单一产地和其它混合产地模型,混合所有4种苹果产地的校正集样本建立的模型取得了最好的预测结果,另外,结合CARS-SPA筛选的16个特征波长,模型得到了进一步简化,其预测相关系数和预测均方根误差分别为0.978和0.441°Brix。结果表明,利用多个产地的苹果样本建立的混合模型,结合有效特征波长,可提高对苹果可溶性固形物含量的预测精度,减小产地差异对可溶性固形物近红外光谱检测的影响。

In order to improve the precision and robustness in determination of soluble solids content （SSC） of ‘ Fuji＇ apple by NIR spectroscopy and eliminate the effect of origin variability on the accuracy of NIR calibration models for the SSC, sample set partitioning based on joint x-y distances （SPXY） was used to select representative subset from the apple samples of 4 different origins. As a comparison, partial least square （PLS） was used to establish local origin and hybrid origin models for the prediction of SSC in apple. Competitive adaptive reweighted sampling （CARS） and successive projections algorithm （SPA） were implemented to select effective variables of the NIR spectroscopy of SSC of apple. The results indicated that the PLS model established based on the 4 origin apple samples performed better than local origin and other hybrid origin models. The model could be effectively simplified using 16 characteristic variables selected by CARS-SPA method from full-spectrum which had 3112 wavelengths. The correlation coefficient （Rp） and root mean square error of prediction （RMSEP） were 0.978 and 0. 441 °Brix, respectively for SSC. It was found that the model developed by more samples of different origins combined with effective wavelengths showed good prediction ability for apple sample of unknown origin, which indicated that it could significantly reduce the origin effect on the robustness of NIR models for SSC of apple.