应用可见-近红外光谱快速识别沙棘汁品牌  

Study on Fast Discrimination of Seabuckthorn Juice Varieties Using Visible-Nir Spectroscopy

作　　者：张海红[1] 张淑娟[1] 王凤花[1] 介邓飞[1] 赵华民[1] 

机构地区：[1]山西农业大学工学院,山西太谷030801

出　　处：《光学学报》2010年第2期574-578,共5页Acta Optica Sinica

基　　金：山西省科技攻关项目(2007031109-2)资助课题

摘　　要：为了实现沙棘汁品牌的快速无损鉴别,提出了采用可见-近红外光谱分析技术(NIR)鉴别沙棘汁品牌的方法。采用FieldSpec3光谱仪对三种沙棘汁进行光谱分析,各获取40个样本数据。采用平均平滑法和多元散射校正(MSC)方法对样本数据进行预处理,再用主成分法(PCA)对光谱数据进行聚类分析并获得各主成分数据。将120个沙棘汁样本随机分成90个建模样本和30个预测样本,把基于累计可信度选择的建模样本的8个主成分(PCs)数据作为BP网络的输入变量,沙棘汁品牌作为输出变量,建立三层反向传播(BP)神经网络鉴别模型,并对30个预测样本进行预测。结果表明,在阈值设定为±0.1的情况下,该模型对预测集样本品牌鉴别准确率达到了100%。所以应用近红外光谱技术结合主成分分析和BP神经网络算法识别沙棘汁品牌是一种有效的方法。In order to achieve non-destructive variety identification of seabuckthorn juice,a fast discrimination method based on visible-near infrared reflectance （NIR） spectroscopy was put forward. A Field Spec 3 spectroradiometer was used for collecting 40 sample spectral data of three varieties of seabuckthorn juice separately. Average smoothing method and multiplicative scattering correction （MSC） method were used to complete the pretreatment of sample data. Then principal component analysis （PCA） was used to process the spectral data after pretreatment. A total of 120 seabuckthorn juice samples were divided into calibration set and validation set randomly,the calibration set had 90 samples and validation set had 30 samples. Eight principal components （PCs） were selected based on accumulative reliabilities which would be taken as the inputs of the three-layer back-propagation neural network,and seabuckthorn juice varieties were selected as the outputs of back propagation （BP） neural network. Then this model was used to predict 30 samples in the validation set. The result showed that a 100% recognition ratio was achieved with the threshold predictive error ±0.1. It could be concluded that PCA combined with BP neural network was an available method for varieties recognition of seabuckthorn juice based on NIR spectroscopy.

关 键 词：可见-近红外光谱 主成分分析 人工神经网络 品牌 沙棘汁 

分 类 号：S123[农业科学—农业基础科学]