欢迎访问《中国农学通报》，
中国农学通报

中国农学通报 ›› 2015, Vol. 31 ›› Issue (6): 241-246.doi: 10.11924/j.issn.1000-6850.casb14110089

所属专题： 烟草种植与生产

• 食品 营养 检测 安全 • 上一篇    下一篇

基于随机森林的烤烟香型分类研究

郭东锋1,胡海洲2,汪季涛1,姚忠达1,杨 辉3,徐 玮3,刘新民2   

  1. (1安徽中烟工业有限责任公司技术中心,合肥 230088；2中国农业科学院烟草研究所,山东青岛 266101；3贵州中烟工业有限责任公司,贵阳 550001）
  • 收稿日期:2014-11-16 修回日期:2015-02-10 接受日期:2015-01-09 出版日期:2015-03-20 发布日期:2015-03-20
  • 通讯作者: 刘新民
  • 基金资助:
    安徽中烟工业有限责任公司科技项目“皖南烟叶烘烤工艺优化与研究”(2013132)；安徽中烟工业有限责任公司科技项目“皖南烟叶生产等级结构优化技术研究”(2014125)。

Study on the Classification of Flue-cured Tobacco Based on the Random Forest Algorithm

Guo Dongfeng1, Hu Haizhou2, Wang Jitao1, Yao Zhongda1, Yang Hui3, Xu Wei3, Liu Xinmin2   

  1. (1Technology Center of Anhui Cigarette Industrial Co. Ltd., Hefei 230088;2Tobacco Research Institute of Chinese Academy of Agricultural Sciences, Qingdao Shandong 266101;3China Tobacco Guizhou Industrial Co., Ltd., Guiyang 550001)
  • Received:2014-11-16 Revised:2015-02-10 Accepted:2015-01-09 Online:2015-03-20 Published:2015-03-20

PDF (PC)

492

摘要： 为找出影响烤烟香型分类的关键因素，以全国6个植烟区域的清、中、浓3个香型烤烟烟叶为研究对象，运用随机森林分类算法对烤烟香型的分类进行系统分析，结果表明：（1）随机森林分类算法对烤烟清、中、浓3种香型整体分类效果正确率达到82.35%，其中对清香型分类效果最好，正确率为100%，对浓香型分类效果一般，中间香型分类效果较差。（2）基于随机森林分类算法输出，影响烤烟香型分类关键的烟叶香味成分主要有：苯甲醇、β-紫罗兰酮、2-环戊烯-1,4-二酮、2-乙酰基-5-甲基呋喃、棕榈酸甲酯、5-羟甲基糠醛、3-羟基-β-大马酮、β-二氢大马酮、糠醛、苯乙醇、二氢猕猴桃内酯等成分。因此，随机森林分类算法能够运用于烤烟香型分类研究中，并在清香型和浓香型分类中取得良好的分类效果，同时找出了能够用于香型分类的关键因素，这对于烤烟原料的深入研究具有一定的参考意义。

关键词: 生物量, 生物量

Abstract: In order to find out the key factors affecting the tobacco flavor classification, 3 types of flue-cured tobacco leaves in 6 tobacco planting areas in China were used as the research objects, and the objects were classified based on the random forest algorithm. The results showed that the correct rate made up 82.35% of the whole samples, in particular the correct rate for QING type reached 100%, but the correct rate for the other types did not match the reality perfectly. Meanwhile the importance of each variable could be reflected under random forest algorithm. In this case the key factors was as following: benzyl alcohol, β-ionone, 2-cyclopentene-1,4-dione, 2-acetyl-5-methyl-furan, methyl palmitate, 5 -(hydroxymethyl) furfural, 3-hydroxy-β-damascone, β-dihydro damascene, furfural, phenylethanol, dihydroactinidiolide etc. Therefore, the random forest algorithm could be applied to the study of tobacco flavor classification, achieve good result in the classification of overall flue-cured tobacco samples, and find out the key factors of classification. Therefore, the random forest algorithm could be explored in the tobacco research.