应用高光谱成像技术鉴别绿茶品牌研究

发布时间：2018-02-24 06:09

本文关键词： 灰度共生矩阵绿茶主成分分析最小二乘支持向量机　出处：《光谱学与光谱分析》2014年05期 　论文类型：期刊论文

【摘要】：应用高光谱成像技术,基于光谱主成分信息和图像信息的融合实现名优绿茶不同品牌的鉴别。首先采集6个品牌名优绿茶在380~1 023nm波长范围的512幅光谱图像,然后提取并分析绿茶样本的可见近红外光谱响应特性,结合主成分分析法找到了最能体现这6类样本差异的2个特征波段(545和611nm),并从这2个特征波段图像中分别提取12个灰度共生矩阵纹理特征参量包括中值、协方差、同质性、能量、对比度、相关、熵、逆差距、反差、差异性、二阶距和自相关,最后融合这12个纹理特征和三个主成分特征变量得到名优绿茶品牌识别的特征信息,利用LS-SVM建立区分模型,预测集识别率达到了100%,同时采用ROC曲线的评估方法来评估分类模型。结果表明综合应用灰度共生矩阵变量和光谱主成分变量作为LSSVM模型输入可实现对绿茶品牌的鉴别。
[Abstract]:Based on the fusion of spectral principal component information and image information, different brands of famous and excellent green tea were identified by hyperspectral imaging technology. Firstly, 512 spectral images of six famous green tea brands were collected in the wavelength range of 380,1023nm. Then we extracted and analyzed the response characteristics of green tea samples by visible and near infrared spectroscopy. Combining principal component analysis (PCA), we find out the two characteristic bands that can best reflect the difference of these six kinds of samples, and extract 12 grayscale co-occurrence matrix texture feature parameters, including median value, covariance, homogeneity, from these two characteristic band images. Energy, contrast, correlation, entropy, inverse gap, contrast, difference, second order distance and autocorrelation. Finally, the 12 texture features and three principal component feature variables are fused to obtain the characteristic information of the famous green tea brand. Using LS-SVM to build a differentiation model, The recognition rate of prediction set is 100 and the classification model is evaluated by ROC curve evaluation method. The results show that green tea brand identification can be realized by using gray co-occurrence matrix variable and spectral principal component variable as LSSVM model input.
【作者单位】：浙江大学生物系统工程与食品科学学院;华东交通大学机电工程学院;
【基金】：国家“十二五”科技计划课题(2011BAD20B12) 国家高技术研究与发展项目(2011AA100705) 中央高校基本科研业务费专项资金资助
【分类号】：O433

【参考文献】