用混合线性分析法建立苹果糖度近红外光谱预测模型

用混合线性分析法的一种变形算法建立了苹果糖度近红外光谱预测模型，并与偏最小二乘模型进行比较。结果表明：虽然最佳的混合线性分析法模型（18个主因子）比最佳偏最小二乘模型（11个主因子）复杂．但其精度却明显优于偏最小二乘模型：利用梗正集的28个苹果样本建立的糖度混合线性分析法校正模型，其相关系数r^2和标准偏差SEC分别为0．92509和0．40618；该校正模型经预测集的11个样本验证，相关系数r^2和标准偏差SEP分别达到0.87611和0．48480。混合线性分析法建立的糖度模型对苹果光谱的校正标准偏差SEC和预测标准偏差SEP分别比PLS法的SEC（0．41473）和SEP（0．50473）减小了2％和3．9％。结果表明：在诸如苹果糖度这一类农产品品质综合指标（非纯组分含量指标）的光谱检测中，应用混合线性分析法进行定量分析是完全可行的。并且其结果可与偏最小二乘法（PLS）的结果相媲美。

Building of Near infrared prediction model of apple sugar content using hybrid linear analysis

A variant algorithm of hybrid linear analysis (HLA/XS) which mostly used in the quantitative analysis of pure components in mixture was first transferred and applied to the near infrared (NIR) determination of sugar content (not a content of pure component) of apples. Results showed that hybrid linear analysis (HLA/XS) produced more precise results than partial least square (PLS),though the optimal HLA/XS model (consisted of 18 principle factors) was more complicate than the optimal PLS model (11 principle factors). The HLA/XS model of sugar content developed from 28 calibration samples gave a very high correlation coefficient (r2) of 0.92509 in calibration set, with the standard error of calibration (SEC) of 0.40618; validated by the 11 prediction samples, the model produced a correlation coefficient (r2) of 0.87611, with a stand error of prediction (SEP) of 0.48480. Compared with the PLS model, HLA/XS decreased SEC and SEP by 2% and 3.9% respectively (the SEC and SEP of PLS model was 0.41473 and 0.50473, respectively). These results indicated that HLA/XS algorithm was completely applicable to quantitative analysis in the spectroscopy determination of integrated index of farm product quality such as sugar content of apples.