两种稻谷胶稠度测定方法的比较分析

为探索快速准确检测稻谷胶稠度的方法,本研究通过近红外漫反射红外光谱技术(NIRDRS)和傅里叶变换中红外漫反射红外光谱技术(FTIRDRS)结合偏最小二乘法(PLS),分别建立107个稻谷样品的胶稠度快速测定红外模型,而后利用区间偏最小二乘法(iPLS)及反向区间偏最小二乘法(BiPLS)对模型进行优化,得到较优的胶稠度测定分析通用模型。结果表明,DRIFTS原始光谱经7点平滑预处理和BiPLS优化,得到最佳模型的交互验证系数(R²)、交叉验证均方差(RMSECV)、预测均方差(RMSEP)及相对分析误差(RPD)分别为0.965 81、4.79、4.73及2.66。最佳近红外漫反射光谱模型是经多元散射校正(MSC)预处理、BiPLS优化后建立的,其R²、RMSECV、RMSEP及RPD分别为 0.964 58、4.35、3.68及3.42。10组外部验证性试验中NIRDRS模型的平均相对误差为1.93%,FTIRDRS模型的平均相对误差为2.60%,表明两种方法均对稻谷胶稠度含量有较强的预测能力和良好的预测效果,均有替代传统国标法测定稻谷胶稠度的潜力。

Comparing FTIRDRS and NIRDRS for the Determinations of Rice Consistency

In order to determinate rice consistency rapidly and accurately, the near infrared diffuse reflection(NIRDRS)and Fourier transform infrared diffuse reflection(FTIRDRS)combined with the partial least squares(PLS) was applied in this paper. Then the models were optimized using interval partial least square(iPLS) and reverse interval partial least square(BiPLS) with 107 rice variety. The results showed that the best near-infrared model for rice consistency was preprocess with 7 smoothing points method and BiPLS as optimizing. The coefficient correlation(R²), cross validation mean variances (RMSECV), root mean square error of prediction (RMSEP) and residual prediction deviation(RPD) were 0.964 58, 4.35, 3.68 and 2.66 respectively. Meanwhile, the best mid-infrared model was obtained by MSC preprocessing method and BiPLS optimizing. The model of R², RMSECV, RMSEP and RPD were 0.965 81, 4.79, 4.73 and 3.42 respectively. In the validation tests, the relative standard deviation of FTIRDRS was about 2.60%, and the relative standard deviation of NIRDRS was about 1.93%. Both models can effectively forecast the consistency of rice grains and have potential to replace national standard method for testing rice consistency.