河南农业科学 ›› 2022, Vol. 51 ›› Issue (10): 161-168.DOI: 10.15933/j.cnki.1004-3268.2022.10.018

• 农业信息与工程·农产品加工 • 上一篇    下一篇

基于轮廓纹理特征和线性判别分析的鲜烟叶部位识别方法

赵晨1,赵浩斌2,路晓崇1,张晓阳3,白涛4,毛岚4,宋朝鹏1,王涛4   

  1. (1.河南农业大学烟草学院,河南 郑州 450002;2.河南省烟草公司,河南 郑州 450018;3.山东中烟工业有限责任公司,山东 济南 250014;4.云南省烟草公司曲靖市公司,云南 曲靖 655000)
  • 收稿日期:2022-05-10 出版日期:2022-10-15 发布日期:2022-12-12
  • 通讯作者: 王涛(1988-),男,福建福州人,农艺师,硕士,主要从事烟叶生产管理与技术推广工作。E-mail:411159344@qq.com
  • 作者简介:赵晨(1998-),女,河南平顶山人,在读硕士研究生,研究方向:烟草调制技术。E-mail:qchen54533@qq.com
  • 基金资助:
    中国烟草总公司科技重点研发项目(110202102007);中国烟草总公司云南省公司资助项目(2021530000241036)

Green Tobacco Position Identification Method Based on Contour‑Texture Features and LDA

ZHAO Chen1,ZHAO Haobin2,LU Xiaochong1,ZHANG Xiaoyang3,BAI Tao4,MAO Lan4,SONG Zhaopeng1,WANG Tao4   

  1. (1.College of Tobacco,Henan Agricultural University,Zhengzhou 450002,China;2.Tobacco Company of Henan
    Province,Zhengzhou 450018,China;3.China Tobacco Shandong Industry Co.,Ltd.,Jinan 250014,China;4.Qujing Tobacco Company of Yunnan Province,Qujing 655000,China)
  • Received:2022-05-10 Published:2022-10-15 Online:2022-12-12

摘要: 为实现鲜烟叶采收部位的数字化识别,进一步提升采收鲜烟叶素质的一致性,利用轮廓纹理特征和线性判别分析(LDA)技术对不同着生部位鲜烟叶进行研究,首先,对采集的鲜烟叶图像进行图像缩放、灰度化、二值化等预处理操作,提取狭长度、矩形度等4个轮廓特征参数,进而提取鲜烟叶图像的灰度共生矩阵(GLCM)特征,并通过LDA进行特征降维,之后利用K近邻算法(KNN)对鲜烟叶部位进行分类。结果表明,所提取未经降维处理的轮廓纹理特征在不同分类模型中的识别准确率均达到0.80以上,可有效反映鲜烟叶部位特征。相对于主成分分析(PCA)处理和未经降维处理,采用LDA降维处理的模型识别准确率最高。所构建的基于KNN算法的鲜烟叶部位识别模型,其精确率、召回率、F1分数、准确率均达到0.99,能够较好地识别鲜烟叶着生部位。

关键词: 鲜烟叶部位, 特征识别, 轮廓纹理特征合, 线性判别分析, 图像识别

Abstract: The objective of this paper was to realize the digital identification of the harvest position of green tobacco leaves,and improve the consistency of the quality of green tobacco leaves harvested.In this paper,tobacco leaves images were taken as the research object,a K‑nearest neighbor algorithm(KNN)recognition method based on contour‑texture characteristics and linear discriminant analysis(LDA)was proposed.Firstly,the pre‑processing operations such as image scaling,grayscale and binarization were performed on the collected green tobacco images,and four contour feature parameters such as narrow length degree,rectangularity were extracted. The gray level co‑occurrence matrix(GLCM) features extracted based on green tobacco images were fused,and the features were reduced by LDA.Finally,the KNN was used to classify green tobacco positions.The recognition accuracy of the extracted contour‑texture features without dimensionality reduction reached more than 0.80 in different classification models,which could reflect the green tobacco leaf position features effectively.The recognition accuracy of the model with LDA dimensionality reduction was the highest compared with that with PCA and without dimensionality reduction.The accuracy,recall rate,F1 score,and precision of the constructed KNN algorithm‑based green tobacco leaf position recognition model all reached 0.99,which can better identify the green tobacco leaf bearing positions.

Key words: Green tobacco leaves position, Feature recognition, Contour?texture feature, Linear discriminant analysis, Image recognition

中图分类号: