【摘要】 应用近红外光谱分析(NIR)技术结合支持向量机(SVM)测定花椒挥发油的含量。以105份样品作为校正集,分别选取epsilon-SVR、nu-SVR两种SVM类型,并采用Linear、Poly、RBF与Sigmoid四种不同核函数进行SVM回归建模,以所建立的校正模型对36份样品的挥发油含量进行预测。结果表明:当SVM类型为epsilon-SVR,核函数为Sigmoid,惩罚参数取109,γ取1×10-6时,所建立的花椒挥发油SVM-NIR模型预测效果最好:R326=0.9317,RMSEP36=0.4268。同时对基于SVM-NIR、PLS-NIR、PCA-BP-NIR和PCA-RBF-NIR的花椒挥发油模型的预测性能进行比较分析,表明SVM-NIR模型具有较强的预测能力(或泛化能力),优于其余3种模型。更多还原

【Abstract】 Grading of zanthoxylum bungeagum maxim needs a method of detecting volatile oil more rapidly and efficiently.The NIR technique and SVM were applied to detecte volatile oil content.Using 105 samples as the calibration set,the models were established by choosing two types of SVM(epsilon-SVR、nu-SVR)and four kinds of kernel function(Linear,Poly,RBF,Sigmoid),respectively.Using the calibration models to predict the volatile oil content from 36 samples,the results showed that the model was the best with penalty factor and γ being 109 and 1×10-6,respectively,when choosing epsilon-SVR as the type of SVM,the sigmoid as kernel function.The determination coefficients(R236)and the root mean square errors of prediction(RMSEP36)were 0.931 7 and 0.426 8,respectively.The models based on SVM-NIR,PLS-NIR,PCA-BP-NIR and PCA-RBF-NIR were compared with each other,and the model based on SVM-NIR was better than it based on the others in respect of generalization.更多还原