基于近红外光谱和最小二乘支持向量机的聚丙烯酰胺类型鉴别

提出了一种基于近红外光谱分析技术和最小二乘支持向量机的鉴别方法，能够快速、无损鉴别聚丙烯酰胺的三种类型。获取非离子，阴离子和阳离子等三种类型的聚丙烯酰胺样本的近红外漫反射光谱，用主成分分析方法对样本光谱数据进行降维，并提取主成分。基于前三个主成分对三种类型的聚丙烯酰胺样本进行聚类分析，并将主成分作为最小二乘支持向量机的输入。通过基于网格搜索的交叉验证方式优化最小二乘支持向量机的参数和作为其输入的主成分个数。每种类型聚丙烯酰胺各采集60个样本，共采集180个样本，每种类型样本随机选取45个样本，共135样本作为训练样本集，剩余45个样本作为测试集。为了验证该方法能否鉴别掺假样本，制备了掺入不同比例非离子聚丙烯酰胺的5个阴离子和5个阳离子聚丙烯酰胺样本。采用基于训练样本集交叉验证预测误差的F统计显著性检验方法来确定样本的鉴别结果误差阈值。结果表明，预测测试集时，准确率为100%。预测10个混和样本时，所有混合样本都被准确识别出。说明该方法能快速无损鉴别不同类型的聚丙烯酰胺并且具有掺假鉴别能力，为聚丙烯酰胺类型的快速鉴别提供了一种新方法。

Discrimination of Types of Polyacrylamide Based on Near Infrared Spectroscopy Coupled with Least Square Support Vector Machine

In this paper, a novel discriminant methodology based on near infrared spectroscopic analysis technique and least square support vector machine was proposed for rapid and nondestructive discrimination of different types of Polyacrylamide. The diffuse reflectance spectra of samples of Non-ionic Polyacrylamide, Anionic Polyacrylamide and Cationic Polyacrylamide were measured. Then principal component analysis method was applied to reduce the dimension of the spectral data and extract of the principal compnents. The first three principal components were used for cluster analysis of the three different types of Polyacrylamide. Then those principal components were also used as inputs of least square support vector machine model. The optimization of the parameters and the number of principal components used as inputs of least square support vector machine model was performed through cross validation based on grid search. 60 samples of each type of Polyacrylamide were collected. Thus a total of 180 samples were obtained. 135 samples, 45 samples for each type of Polyacrylamide, were randomly split into a training set to build calibration model and the rest 45 samples were used as test set to evaluate the performance of the developed model. In addition, 5 Cationic Polyacrylamide samples and 5 Anionic Polyacrylamide samples adulterated with different proportion of Non-ionic Polyacrylamide were also prepared to show the feasibilty of the proposed method to discriminate the adulterated Polyacrylamide samples. The prediction error threshold for each type of Polyacrylamide was determined by F statistical significance test method based on the prediction error of the training set of corresponding type of Polyacrylamide in cross validation. The discrimination accuracy of the built model was 100% for prediction of the test set. The prediction of the model for the 10 mixing samples was also presented , and all mixing samples were accurately discriminated as adulterated samples. The overall results demonstrate that the discrimination method proposed in the present paper can rapidly and nondestructively discriminate the different types of Polyacrylamide and the adulterated Polyacrylamide samples, and offered a new approach to discriminate the types of Polyacrylamide.