基于近红外漫反射光谱和PCA-SVM算法快速鉴别炉甘石

Rapid Identification of Calamina Based on Near-infrared Diffuse Reflectance Spectroscopy and PCA-SVM Algorithm

目的:利用主成分分析(PCA)和支持向量机(SVM)算法,建立炉甘石生品、伪品及炮制品的近红外漫反射光谱(NIRS)鉴别模型。方法:采集炉甘石生品、伪品及炮制品的NIRS,选取特征谱段,优选光谱预处理方法及最佳主成分数,建立PCA-SVM鉴别模型。结果:在7 500~4 000 cm-1谱段,以一阶导数法(FD)为最佳光谱预处理方法,PCA提取的光谱前5个主成分为最佳主成分,并经网格搜索算法确定惩罚因子c=0. 25,核函数参数g=8为最佳SVM内部参数,建立炉甘石PCA-SVM鉴别模型。该模型五折交叉验证准确率100%,且模型对训练集和测试集样品预测正确率亦均达100%。结论:基于PCASVM算法所建立的炉甘石NIRS鉴别模型预测准确率高,结合固体粉末漫反射技术无损、快速的优点,该模型可用于炉甘石生品、伪品及炮制品的无损、快速鉴别。

Objective:To establish a near-infrared diffuse reflectance spectroscopy(NIRS identification model for crude products,counterfeit products and processed products of Calamina by principal component analysis(PCA)and support vector machine(SVM)algorithm.Method:NIRS of crude products,counterfeit products and processed products of Calamina were collected,the characteristic spectrum segments were selected,the preprocessing method and the optimum principal component number were optimized,and the PCASVM qualitative model was established.Result:The characteristic spectrum segment of analysis model was 7 500-4 000 cm-1.Spectra were preprocessed by the first-order derivative method(FD).The optimum principal component number was 5.And the optimum internal parameters of SVM[penalty factor(c)=0.25 and kernel function parameter(g)=8]were screened by applying the grid search algorithm.In the PCA-SVM qualitative model,the prediction accuracy rate was 100%for the 5-fold cross validation,and the prediction accuracy rates also were 100%both for training set and test set.Conclusion:PCA-SVM analysis model of NIRS for Calamina samples has a high prediction accuracy rate,and it can be used for the rapid and nondestructive identification of crude products,counterfeit products and processed products of Calamina by combining the diffuse reflection technique on solid powder.