基于连续小波变换的烤烟叶片总氮和烟碱含量高光谱监测预测研究  

Hyperspectral monitoring and prediction of total nitrogen and nicotine contents in flue-cured tobacco leaves based on continuous wavelet transform

作　　者：蒋薇 汤亮[2] 薛博文 程涛[2] 孙伟超 李军营[1] JIANG Wei;TANG Liang;XUE Bowen;CHENG Tao;SUN Weichao;LI Junying(Yunnan Academy of Tobacco Agriculture Science,Kunming 650021,China;National Engineering and Technology Center for Information Agriculture,Nanjing Agricultural University,Nanjing 210095,China;Aerospace Information Research Institute,Chinese Academy of Sciences,Haidian District 100094,China)

机构地区：[1]云南省烟草农业科学研究院,云南650021 [2]南京农业大学国家信息农业工程技术中心,江苏210095 [3]中国科学院空天信息创新研究院,北京100094

出　　处：《中国烟草学报》2023年第4期33-43,共11页Acta Tabacaria Sinica

基　　金：中国烟草总公司云南省公司科技计划重大项目“基于高光谱的烟叶产量预估和品质监测关键技术研究与应用”(No.2021530000241025)。

摘　　要：【目的】明确关键生育期及烤后烟叶总氮和烟碱含量的光谱响应规律并构建监测预测模型,以期及时准确地获取烟叶总氮及烟碱含量。【方法】通过实施不同品种、不同施氮水平互作处理的烤烟田间试验,以不同处理的植株冠层高光谱为数据源,通过多尺度连续小波变换和敏感小波特征提取,构建各生育期及烤后烟叶总氮和烟碱的监测预测模型。【结果】(1)不同生育期鲜烟叶总氮及烟碱含量的最优小波特征有所差异:总氮的最优小波特征位置随着生育进程的延长向长波方向移动,从427nm(旺长期)到720 nm(打顶期)再到769 nm(中部叶成熟期);而烟碱在打顶期(WF_(715,6))和中部叶成熟期(WF_(714,3))具有较为一致的特征位置,但最优尺度不同,因此在不同生育期分别建立监测模型是必要的;(2)打顶期的总氮及烟碱含量监测精度较高,该时期的敏感小波特征WF_(720,7)与烟株叶片的总氮含量相关系数较大(r=-0.83);特征WF_(715,6)与烟株叶片的烟碱含量相关系数较大(r=-0.81),且模型验证集NRMSE均在20%以内;(3)打顶期对烤后烟叶总氮和烟碱含量预测精度优于旺长期及中部叶成熟期,使用打顶期冠层光谱提取的敏感小波特征WF_(628,3)预测总氮含量的NRMSE为7.37%,WF_(856,3)预测烟碱含量的NRMSE为12.6%。【结论】本研究基于敏感小波特征构建的各生育期烤烟叶片总氮及烟碱含量监测和预测模型具有较好的准确性,为烤烟氮素管理调控、品质监测预测以及烤烟加工分级等提供支持。[Background]This study aims to clarify the spectral response patterns of total nitrogen and nicotine contents in tobacco leaves during critical growing stages and after-curing stages,and construct a monitoring and prediction model,in order to obtain timely and accurate total nitrogen and nicotine contents in tobacco leaves.[Methods]A field experiment involving different tobacco varieties under nitrogen application levels was conducted,and the plant canopy hyperspectral data of different treatments was obtained.Through multi-scale continuous wavelet transform and extraction of sensitive wavelet features,a monitoring and prediction model was established for total nitrogen and nicotine contents in in tobacco leaves at different growth stages and after-curing stage.[Results](1)The optimal wavelet characteristics of total nitrogen and nicotine contents in fresh tobacco differed at different growth stages.The optimal wavelet feature position of total nitrogen in fresh tobacco leaves shifted towards the long-wave direction as the growth process progresses,from 427 nm(peak stage)to 720 nm(topping stage)to 769 nm(at middle leaf maturity stage);while nicotine had more consistent feature positions at topping stage(WF_(715,6))and middle leaf maturity stage(WF_(714,3)),but with different optimal scales.Therefore,it is necessary to establish separate monitoring models for different growth stages.(2)The accuracy of monitoring total nitrogen and nicotine contents in the topping period was higher than other periods.The sensitive wavelet feature WF_(720,7) had a biggest correlation coefficient with the total nitrogen content of tobacco leaves(r=-0.83).The sensitive wavelet feature WF_(715,6) was highly correlated with the nicotine content of tobacco leaves(r=-0.81),and the NRMSE of the model validation set was within 20%.(3)The canopy spectra of topping stage predicted the total nitrogen and nicotine contents in flue-cured tobacco better than the long prosperous stage and the middle leaf maturity stage.The sensitive wavelet featur

关 键 词：反射光谱 小波分析 鲜烟叶 烤后烟叶 品质预测 

分 类 号：O657.3[理学—分析化学] TS411[理学—化学]