基于BPNN的烤烟褐变程度识别及其分类烟叶质量分析  

Identification of tobacco leaf browning degree based on BPNN and quality analysis of classified tobacco leaves

作　　者：孙光伟 郭青青 柳均[3] 冯吉 孙敬国 张鹏龙 吴哲宽 李建平 陈振国 SUN Guang-wei;GUO Qing-qing;LIU Jun;FENG Ji;SUN Jing-guo;ZHANG Peng-long;WU Zhe-kuan;LI Jian-ping;CHEN Zhen-guo(Hubei Province Tobacco Research Institute,Wuhan 430030,China;College of Life Sciences,Hubei University,Wuhan 430062,China;Hubei Province Tobacco Quality Supervision&Test Station,Wuhan 430030,China;Enshi Branch of Hubei Province Tobacco Company,Enshi,Hubei 445000,China)

机构地区：[1]湖北省烟草科学研究院,武汉430030 [2]湖北大学生命科学学院,武汉430062 [3]湖北省烟草公司烟草质量监督检测站,武汉430030 [4]湖北省烟草公司恩施州公司,湖北恩施445000

出　　处：《西南农业学报》2023年第1期169-177,共9页Southwest China Journal of Agricultural Sciences

基　　金：中国烟草总公司科技项目(110202102007);湖北省烟草公司重点科技项目(027Y2021-005)。

摘　　要：【目的】建立BP(Back propagation)神经网络(BPNN)自动识别系统,以实现烤烟褐变标准化和量化。【方法】以云烟87上部烟叶为样本,通过扫描获取烟叶颜色等特征信息,建立BP神经网络烤烟褐变程度识别模型,输出判别结果,以人工判别烤烟褐变结果为参考,进行相似度比较。通过外观质量、常规化学成分、多酚含量、TSNAs含量和感官质量变化分析比对,验证BP神经网络自动识别系统和人工识别结果的精准度。【结果】建立的BP神经网络模型能够精准识别烟叶褐化等级,其识别准确率为98.75%,分级烟叶外观质量、常规化学成分、多酚含量、TSNAs含量和感官质量变化与人工识别基本一致,两种识别模式无显著性差异(P<0.05),为杂色烟分级提供了客观评价方法,能有效区分不同褐变程度烟叶的可用性。【结论】BP神经网络识别系统对烟叶褐变程度鉴别精准度与人工识别接近,可以利用BP神经网络自动识别系统替代人工识别。【Objective】The present paper aimed to establish a BP(Back propagation)neural network(BPNN)automatic recognition system to achieve the standardization and quantification of flue-cured tobacco browning.【Method】Taken the upper tobacco leaves of Yunyan 87 as a sample,scanning to obtain characteristic information such as tobacco leaf color,established a BP neural network to identify the degree of browning of flue-cured tobacco,and output the discrimination results,and compared the similarity with the results of artificially discriminating the browning of flue-cured tobacco as a reference.Through the analysis and comparison of appearance quality,conventional chemical composition,polyphenol content,TSNAs content and sensory quality changes,the accuracy of the BP neural network automatic recognition system and manual recognition results were veriofied.【Result】Establishing the BP neural network model could accurately identify tobacco browning level,and its recognition accuracy was 98.75%.There was no significant different among grading tobacco appearance quality,routine chemical compositions,polyphenol content,the content of TSNAs and sensory quality of change and the artificial recognition(P<0.05).Which provided objective evaluation method for noise smoke,and the availability of tobacco leaves with different browning degrees was effectively distinguished.【Conclusion】The accuracy of the BP neural network recognition system for identifying the degree of tobacco browning is close to that of manual recognition.The BP neural network automatic recognition system can be used to replace manual recognition,which provides a reference for promoting the construction of tobacco BP.

关 键 词：烤烟 酶促褐变 褐变程度 BP神经网络 外观质量 

分 类 号：TS44[农业科学—烟草工业]