烤烟的近红外光谱检测模型

NIR Spectral Detection Model Value of the Flue-cured Tobacco

对杀青烟叶蛋白质和水溶性总糖含量的光谱检测,发现近红外光谱(1100—2500nm)的检测模型优于可见-近红外光谱(350—2526nm),烟粉检测模型优于片状烟叶的检测模型。通过对烟叶全部光谱数据不同的预处理来探究其蛋白质和水溶性总糖的近红外光谱的检测模型,并利用近红外有效波长对施木克值的含量进行预测。利用偏最小二乘法(PLS)通过训练集的交叉验证建立回归模型,结果表明:(1)对原始光谱进行二阶导数变换后,得到蛋白质含量预测模型的预测集r=0.9768、RMSE=0.6843;(2)对原始光谱每隔51个点进行移动平滑处理及主成分数为8时,水溶性总糖含量预测模型的预测集r=0.9495、RMSE=0.9049;(3)基于82个波长对施木克值的预测模型的预测集r=0.9356、RMSE=0.1060。

The spectrum detection models of protein and .total soluble sugar contents were refined from the green-removing tobacco. Results show that the NIR （1100-2500nm） better than the V-NIR （350-2526nm） and the powdery than the flaky. Preprocessing the spectrum data through different methods were explored that the NIR spectral detection model of protein and total saccharine contents. Meanwhile the effective wavelengths were built the regression model of the Shmuck Value. PLS and full cross validation were used to build the regression models. And the results show：After 2nd derivative disposal,the model of protein content have the regression coefficient r=0. 9768,RMSE=0. 6843 in prediction set ;when the number of PCs is 8 and smoothing every 51 numbers,the model of total soluble sugar contents have the regression coefficient r=0. 9495,RMSE= 0. 9049 ;Regression model of the Shmuck Value bases on the 82 segments effective wavelengths,which has the regression coefficient r=0. 9356, RMSE = 0. 1060.