摘要
为提高鲜烟叶成熟度的识别精度,提出基于近红外光谱和图像识别的多源信息融合技术的烟叶成熟度判别方法:利用随机森林(RF)方法分别建立近红外光谱判别模型、图像判别模型和多源信息融合判别模型,对烟叶成熟程度进行检测。近红外光谱模型对红花大金元、K326和云烟87等3个烤烟品种烟叶成熟度的识别正确率分别为91.27%、90.43%、89.44%,图像模型的识别正确率分别为86.20%、86.96%、81.23%,融合模型的识别正确率分别为94.08%、94.78%和92.96%。与近红外光谱模型相比,融合模型的判别正确率平均提高了3.93%;与图像模型相比,融合模型的判别正确率平均提高了10.83%。
In order to improve the identification accuracy of fresh tobacco leaf maturity,a tobacco leaf maturity identification method was proposed based on the multi-source information fusion technology of near infrared spectrum and image recognition.Using random forest method,the near infrared spectrum discriminant model,image discriminant model and multi-source information fusion discriminant model were established to detect and analyze the maturity of tobacco leaves.The accuracy rates of the near infrared spectrum model for the three flue-cured tobacco leaf maturity were 91.27%,90.43%and 89.44%,and the accuracy rates of the image model for the three flue-cured tobacco leaf maturity were 86.20%,86.96%and 81.23%,respectively.The accuracy of the fusion model for the identification of leaf maturity of the three flue-cured tobacco varieties was 94.08%,94.78%and 92.96%,respectively.Compared with near infrared spectroscopy model and image model,the accuracy of fusion model is improved by 3.93%and 10.83%on average.
作者
杨睿
宾俊
苏家恩
汪华国
王文伦
何承刚
陈颐
邹聪明
YANG Rui;BIN Jun;SU Jiaen;WANG Huaguo;WANG Wenlun;HE Chenggang;CHEN Yi;ZOU Congming(Yunnan Academy of Tobacco Agricultural Sciences,Kunming,Yunnan 650031,China;College of Tobacco,Yunnan Agricultural University,Kunming,Yunnan 650201,China;College of Tobacco,Guizhou University,Guiyang,Guizhou 550025,China;Dali Tobacco Company of Yunnan,Dali,Yunnan 671000,China;Chuxiong tobacco company of Yunnan,Chuxiong,Yunnan 675005,China)
出处
《湖南农业大学学报(自然科学版)》
CAS
CSCD
北大核心
2021年第4期406-411,418,共7页
Journal of Hunan Agricultural University(Natural Sciences)
基金
中国烟草总公司重点项目(110202102007)
云南省烟草专卖局重点项目(2019530000241019)。
关键词
烟叶成熟度
近红外光谱
图像识别
数据融合
判别
tobacco maturity
near infrared spectroscopy
image recognition
data fusion
discriminant analysis