基于LS-SVM的烤烟烟叶产地判别

为了探索一种快速有效的烤烟烟叶产地鉴别方法,利用近红外光谱技术结合最小二乘支持向量机(LS-SVM)对烤烟烟叶的产地进行了判别。选择云南、湖北、河南三地不同等级烤烟烟叶作为研究对象,对原始光谱数据进行平滑和附加散射校正(MSC)预处理后再进行主成分分析,选择4～12个主成分作为输入变量进行LS-SVM建模。结果显示,该LS-SVM模型预测效果较好,预测相关系数rp≥0.990 7,预测标准误差(SEP)和预测均方根误差(RMSEP)分别为1.755 1和1.737 3,优于偏最小二乘回归(PLS)的预测结果,基于LS-SVM的近红外光谱技术能够很好地对烟叶产地进行判别。

Identification of Producing Area of Tobacco Leaf Based on LS-SVM

In order to explore a fast and efficient method which determines the producing area of tobacco leaf,near-infrared reflectance spectroscopy with least squares-support vector machines(LS-SVM) was applied to determine producing area of tobacco leaf.Three producing areas including Yunnan,Hubei and Henan were selected as the research objects.As the pretreatments of the optimal smoothing way,moving average with three segments and multiplication scatter correction(MSC) were applied to reduce the noise of the spectra.After the principle component analysis,4 to 12 principal components(PCs) were chosen as the inputs of LS-SVM models.The Results show that the prediction performance of the LS-SVM model with 12 PCs is better than partial least square(PLS) model.Its correlation coefficient of prediction set(rp) is 0.990 7,standard error of prediction(SEP) is 1.755 1,and root mean square error of prediction(RMSEP) is 1.737 3.It is concluded that NIR spectroscopy with LS-SVM is a feasible method to determine the producing area of tobacco leaf.