Millet Origin Identification Model Based on Near-infrared Spectroscopy

[Objectives]This study was conducted to clarify the difference of millet from different producing areas in near-infrared spectroscopy(NIRS)modeling.[Methods]Millet samples from six different regions were collected for NIRS analysis,and an origin identification model based on BP neural network was established.The competitive adaptive reweighted sampling(CARS)algorithm was used to extract characteristic wavelength variables,and a CARS-BP model was established on this basis.Finally,the CARS-BP model was compared with support vector machine(SVM),partial least squares discriminant analysis(PLS)and KNN models.[Results]The characteristic wavelengths were extracted by CARS,and the number of variables was reduced from 1845 to 130.The discrimination accuracy of the CARS-BP model for the samples from six producing areas reached 98.1%,which was better than SVM,PSL and KNN models.[Conclusions]NIRS can quickly and accurately identify the origin of millet,providing a new method and way for the origin identification and quality evaluation of millet.

基于神经网络的小米产地鉴别研究

小米的品质与产地息息相关,产地不同可能导致小米品质存在差异。为了实现小米产地的快速、精确鉴别,保护优质小米的品牌效益,以6种不同产地的小米为研究对象,将近红外光谱分析技术与反向传播(Back-propagation,BP)神经网络相结合建立小米产地鉴别模型,使用竞争自适应重加权采样(Competitive Adaptive Reweighted Sampling,CARS)算法提取特征波长变量,并在此基础上建立CARS-BP模型,之后将CARS-BP模型与全谱BP神经网络模型、支持向量机(Support Vector Machine,SVM)、偏最小二乘法(Partial Least Square,PLS)、K最近邻(K-Nearest Neighbor,KNN)分类算法进行比较,对比5种模型鉴别的准确率。结果表明:CARS-BP模型对6种产地小米样品的产地鉴别平均准确率达98.1%,优于SVM、PSL和KNN模型。