应用近红外光谱和多算法融合方法分析烤烟的香型风格特征

发布时间：2018-08-16 17:09

【摘要】：以全国17个主要烤烟产地省份中收集的3 914个烟叶样品的近红外光谱为实验对象,其中浓香型、中间香型、清香型烟叶光谱分别865条、1 403条、1 646条,应用近红外光谱和多算法融合方法分析其香型风格特征。在以产地进行初步划分烟叶香型以及认可过渡型和非典型香型类型的基础上,选取基于主成分及Fisher准则的投影法(PPF)、偏最小二乘判别(DPLS)、支持向量机(SVM)作为单分类器,得到各个算法第1和2判别分析结果;应用PPF-DPLS-SVM融合和各算法第1和2判别分析结果,将预测验证样品的分析结果详细划分为典型、过渡型、非典型香型样品(分别为493,392,115个);其中典型香型烟叶样品的判别准确率达到92.7%,较未进行典型样品划分时PPF,DPLS,SVM单算法的识别准确率分别提高了30.2%,15.4%,16.6%。样品数据来源于全国主要烤烟产地,数据量大,代表性较好,分析结果具有一定普遍性;提出的多算法融合分析方法大幅度提高了通过客观数据判别烤烟香型的准确率;同时,将烤烟香型细划分为典型、过渡型和非典型香型的方式,对烤烟烟叶原料的科学合理利用以及烟叶原料的模块化工业加工等有指导作用。
[Abstract]:The near infrared spectra (NIR) of 3 914 tobacco samples collected from 17 provinces of main flue-cured tobacco producing areas in China were used as experimental objects. Among them, 865 pieces of tobacco leaves were of Luzhou-flavor type, 1 403 pieces of pure aroma type, and 1 646 of pure aroma type tobacco leaves. Near infrared spectroscopy (NIR) and multi-algorithm fusion method are used to analyze its flavor style. On the basis of the primary classification of tobacco leaf aroma type and the approved transition type and atypical flavor type based on the production area, the projection method based on principal component and Fisher criterion, (PPF), partial least square method, was selected to distinguish (DPLS), support vector machine (SVM) as a single classifier. The first and second discriminant analysis results of each algorithm are obtained, and the results of PPF-DPLS-SVM fusion and the first and second discriminant analysis of each algorithm are used to divide the analysis results of the prediction and verification samples into typical and transitional types. The discriminant accuracy of typical aroma tobacco samples was 92.7%, which was higher than that of PPFN DPLSS-SVM single algorithm with 30.2% 15.4% 16.6% higher than that without typical samples classification. Sample data from the main flue-cured tobacco production areas in China, data volume is large, representative, analysis results have a certain universality; the proposed multi-algorithm fusion analysis method greatly improved the accuracy of identification of flue-cured tobacco flavor type through objective data; at the same time, The fine classification of flue-cured tobacco flavor into typical, transitional and non-typical flavor type has a guiding effect on the scientific and rational utilization of flue-cured tobacco leaf raw materials and the modular industrial processing of flue-cured tobacco raw materials.
【作者单位】：中国农业大学信息与电气工程学院;中国农业大学教育部现代精细农业系统集成研究重点实验室;上海烟草集团有限责任公司;
【基金】：国家重点研发计划3课题(2016YFD0700304) 上海烟草集团科技项目资助
【分类号】：O657.33;TS411

【相似文献】