首页 | 本学科首页   官方微博 | 高级检索  
检索        

基于NIRS技术和PCA-SVM算法快速鉴别国产和进口啤酒花
引用本文:郭云香,陈龙,李晓瑾,王果平,蒋益萍,辛海量,贾晓光.基于NIRS技术和PCA-SVM算法快速鉴别国产和进口啤酒花[J].药学实践杂志,2019,37(4):322-331.
作者姓名:郭云香  陈龙  李晓瑾  王果平  蒋益萍  辛海量  贾晓光
作者单位:新疆医科大学 中医学院, 新疆 乌鲁木齐 830011;新疆维吾尔自治区中药民族药研究所, 国家中医药管理局新疆中药民族药资源重点实验室, 新疆 乌鲁木齐 830002;海军军医大学药学院生药学教研室, 上海 200433,襄阳市中心医院 湖北文理学院附属医院, 湖北 襄阳 441021;湖北中医药大学 中药资源和中药复方教育部重点实验室, 湖北 武汉 430065,新疆医科大学 中医学院, 新疆 乌鲁木齐 830011;新疆维吾尔自治区中药民族药研究所, 国家中医药管理局新疆中药民族药资源重点实验室, 新疆 乌鲁木齐 830002,新疆维吾尔自治区中药民族药研究所, 国家中医药管理局新疆中药民族药资源重点实验室, 新疆 乌鲁木齐 830002,海军军医大学药学院生药学教研室, 上海 200433,海军军医大学药学院生药学教研室, 上海 200433,新疆医科大学 中医学院, 新疆 乌鲁木齐 830011;新疆维吾尔自治区中药民族药研究所, 国家中医药管理局新疆中药民族药资源重点实验室, 新疆 乌鲁木齐 830002
基金项目:国家自然科学基金(U1603283)
摘    要:目的 利用近红外漫反射光谱(near-infrared reflectance spectroscopy,NIRS)法,结合主成分分析(principal component analysis,PCA)和支持向量机(support vector machine,SVM)联用算法,建立PCA-SVM的NIR模式识别模型,用于国产和进口啤酒花的快速鉴别。方法 收集上述不同产地的啤酒花样品,制备成均匀粉末,在4 000~12 500 cm-1光谱区,采集各样品粉末的NIR光谱,选取特征谱段9 000~4 100 cm-1为建模谱段,分别采用不同光谱预处理方法进行预处理并分别进行PCA降维。根据2维主成分平面散点图,优选最佳预处理方法。利用最佳预处理方法处理后的光谱PCA降维数据,建立SVM模式识别模型,SVM模型参数采用网格搜索法、遗传算法(GA)、粒子群优化法(PSO)进行寻优。对比不同主成分数所建PCA-SVM模型的预测准确率,确定最佳的主成分数,最终建立PCA-SVM的NIR快速鉴别模型。结果 在6500~5400 cm-1谱段,以一阶导数法(first derivative,FD)为最佳光谱预处理方法,PCA提取的光谱前8个主成分为最佳主成分,并经网格搜索法确定最佳SVM内部参数:惩罚因子c=2,核函数参数g=1,建立啤酒花PCA-SVM鉴别模型,该模型五折交叉验证准确率达97.37%,对校正集和测试集样品预测准确率均分别为97.37%和97.44%。结论 啤酒花NIRS光谱,进行PCA-SVM算法建模,模型预测准确率高、性能佳,可用于啤酒花样品的快速、无损鉴别。

关 键 词:啤酒花  近红外漫反射光谱  主成分分析  支持向量机  定性鉴别
收稿时间:2019/1/9 0:00:00
修稿时间:2019/5/14 0:00:00

Rapid identification of domestic and imported hops based on NIRS technology and PCA-SVM algorithm
GUO Yunxiang,CHEN Long,LI Xiaojin,WANG Guoping,JIANG Yiping,XIN Hailiang and JIA Xiaoguang.Rapid identification of domestic and imported hops based on NIRS technology and PCA-SVM algorithm[J].The Journal of Pharmaceutical Practice,2019,37(4):322-331.
Authors:GUO Yunxiang  CHEN Long  LI Xiaojin  WANG Guoping  JIANG Yiping  XIN Hailiang and JIA Xiaoguang
Abstract:Objective To develop a rapid identification method for domestic and imported hops by the establishment of PCA-SVM model using near-infrared reflectance spectroscopy (NIRS),combined with principal component analysis (PCA) and support vector machine (SVM) algorithm.Methods The hop samples from different sources were collected and ground into uniform powder.The NIR spectra of each powder sample were collected in the range of 4000~12500 cm-1.The characteristic spectrum segment was selected from 9000~4100 cm-1,which was pretreated by different spectral pretreatment methods and subjected to PCA dimensionality reduction.According to the 2-dimensional principal component plane scatter plot,the pretreatment method was optimized.The SVM pattern recognition model was established by using the best preprocessing method to process the PCA dimensionality reduction data of the post-spectrum.The SVM model parameters were searched by grid search method,genetic algorithm (GA) and particle swarm optimization (PSO).The prediction accuracy of the PCA-SVM models built by different principal component numbers were compared to determine the optimal principal component number.Finally,the rapid NIR identification model of PCA-SVM is established.Results In the 6500~5400 cm-1 spectral segment,the first derivative (FD) is the best spectral pretreatment method,and the first 8 principal components are the best principal components of the spectrum extracted by PCA.The optimal SVM internal parameters are determined by the grid search method:the penalty factor(c)=2,the kernel function parameter(g)=1.The prediction accuracy rate of this hop PCA-SVM identification model was 97.37% for the 5-fold cross validation,97.37% for the calibration set and 97.44% for test set samples.Conclusion This model has high accuracy and consistent performance.It can be used for rapid and non-destructive identification of hop samples.
Keywords:Hop  near-infrared diffuse reflectance spectroscopy  principal component analysis  support vector machine  qualitative identification
点击此处可从《药学实践杂志》浏览原始摘要信息
点击此处可从《药学实践杂志》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号