基于近红外光谱与支持向量机的甘薯粉丝掺假快速检测  

Rapid detection of adulterated sweet potato starch noodle by near-infrared spectroscopy and support vector machine

作　　者：陈嘉[1] 高丽[1] 叶发银[1] 雷琳[1] 赵国华[1,2,3] CHEN Jia;GAO Li;YE Fayin;LEI Lin;ZHAO Guohua(College of Food Science,Southwest University,Chongqing 400715,China;Chongqing Sweet Potato Engineering and Technology Centre,Chongqing 400715,China;Chongqing Key Laboratory of Agricultural Product Processing,Chongqing 400715,China)

机构地区：[1]西南大学食品科学学院,重庆400715 [2]重庆市甘薯工程技术研究中心,重庆400715 [3]重庆市农产品加工技术重点实验室,重庆400715

出　　处：《食品与发酵工业》2019年第11期211-218,共8页Food and Fermentation Industries

基　　金：中央高校基本业务费专项资金资助(XDJK2018C014);重庆市社会事业与民生保障科技创新专项项目(cstc2015shms-ztzx80006);广西农产资源化学与生物技术重点实验室开放基金资助项目(KF01)

摘　　要：探索了近红外光谱(near infrared spectra,NIRS)结合支持向量机(support vector machine,SVM)检测甘薯粉丝掺假(掺杂木薯淀粉和玉米淀粉)的可行性。以掺假甘薯粉丝为研究对象,建立了基于NIRS及SVM的甘薯粉丝掺假定性判别及定量分析模型,并通过光谱预处理及光谱变量筛选对模型进行了优化。结果显示,采用标准正态变量变换和一阶导数对全光谱预处理后,甘薯粉丝掺假SVM定性判别模型的识别准确率可达100%,优于马氏距离判别模型;用标准正态变量变换和一阶导数对光谱预处理,并通过前向区间支持向量机(forward interval support vector machine,fi-SVM)筛选光谱变量后,木薯淀粉含量SVM预测模型的相关系数( r )和预测均方差(RMSEP)可达到0.92和11.20,玉米淀粉含量SVM预测模型的 r 和RMSEP可达到0.96和7.49。结果表明,基于NIRS和SVM的甘薯粉丝掺假定性判别及定量分析检测模型具有较高的识别率和预测精度,用于检测甘薯粉丝的掺假是可行的。The aim of this study was to explore the feasibility of identifying and quantifying adulterated sweet potato starch noodles (adulterated with cassava starch and corn starch) using near-infrared spectroscopy (NIRS) and support vector machine (SVM). Qualitative discrimination and quantitative analysis models based on NIRS and SVM were built,which were optimized by spectra pretreatment and spectral variables selection. The results showed that after using standard normal variate transformation and first derivative pretreatment,the accuracy of SVM model for qualitative discrimination based on whole NIRS spectra for identifying adulterated sweet potato starch noodles achieved 100%,which surpassed the Mahalanobis distance discriminant model. Moreover,by using a forward interval support vector machine (fi-SVM) algorithm to screen out spectral variables,the correlation coefficients ( r ) of the SVM models for cassava starch content and corn starch content reached 0.92 and 0.96,respectively. Besides,the root mean square error of prediction (RMSEP) of these two models reached 11.2 and 7.49,respectively. The results indicated that models based on NIRS and SVM for qualitative discrimination and quantified analysis for adulterated sweet potato starch noodles had high recognition rates and prediction accuracy. Therefore,it is feasible to detect adulteration of sweet potato starch noodles by using NIRS and SVM.

关 键 词：甘薯粉丝 掺假 近红外 支持向量机 定性判别 定量分析 

分 类 号：TS237[轻工技术与工程—粮食、油脂及植物蛋白工程]