[目的]研究不同成熟度上部烟叶的高光谱特征及智能化判别的可行性.[方法]运用便携式高光谱仪采集3种成熟度(尚熟(SS)、成熟(CS)、过熟(GS))上部烟叶的高光谱图像并提取光谱数据.运用相关性分析、主成分分析以及方差分析等方法分析光谱特征并构建5种模型(支持向量机(SVM)、K近邻(KNN)、随机森林(RF)、LightGBM和XGBoost)用于成熟度判别评价.[结果](1)可见光(400～720nm)与近红外(750～1000nm)内部各波段之间相关性较强,而两个区域之间相关性较弱;(2)5个特征值大于1的主成分(PC1～PC5)几乎包含了所有的光谱信息,且主成分方差分析结果表明不同成熟度上部烟叶的光谱反射特征在可见光、红边以及部分近红外区域(950～1000 nm)统计学差异显著;(3)5种模型中SVM性能最优,2021年度样品的判别精确率、召回率和F1分数均在0.95以上,而2022年度以及2021+2022年度样品分别在0.93和0.92以上.[结论]上部烟叶高光谱存在多重共线性,具有很好的降维效果,且不同成熟度的光谱反射特征存在显著差异.SVM判别性能在不同年度间具有很好的稳定性,可用于上部烟叶成熟度判别.
[Objective]The purpose of this study was to investigate the hyperspectral characteristics of upper tobacco leaves at different-maturity levels and the feasibility of intelligent discrimination.[Methods]In this study,a portable hyperspectral instrument was used to collect the hyperspectral imagines of upper tobacco leaves of three different-maturity levels(pre-maturity(SS),maturity(CS)and post-maturity(GS))and extracted their spectral data.Their spectral characteristics were studied by using correlation analysis,principal components analysis and variation analysis,and 5 models(SVM,KNN,RF,LightGBM and XGBoost)were constructed for evaluating their discriminant performances of tobacco leaf maturity.[Results]The results showed that:(1)there was a strong correlation among the bands within the visible light(400-720 nm)or the near infrared(750-1000 nm)regions,while the correlation between the two regions was weak.(2)the 5 principal components(PC1-PC5)with eigenvalues greater than 1 almost contained all the hyperspectral information.The spectral reflectance characteristics of upper tobacco leaves with different maturity levels showed significant difference in visible light,red edge and part of near infrared region(950-1000 nm).(3)Among the 5 models,SVM has the best evaluation,with precision,recall and Fl scores for the samples in 2021 above 0.95,and for the samples in 2022 and 2021+2022 above 0.93 and 0.92 respectively.[Conclusion]The hyperspectral data of the upper tobacco leaves exhibit multicollinearity,which has excellent dimensionality reduction effects.Moreover,there are significant differences in spectral reflectance characteristics at different maturity levels.The SVM discriminant performance has good stability across different years and can be used for determining the maturity of upper tobacco leaves.
邓建强;王大彬;乾艳;尹忠春;彭五星;李富强;任晓红
湖北省烟草公司恩施州公司,湖北恩施 445000中国农业科学院烟草研究所,农业农村部烟草质量安全风险评估实验室青岛 266101
便携式高光谱仪;上部烟叶;成熟度;光谱特征;模型构建及应用
portable hyperspectral imager;upper tobacco leaf;maturity;spectral characteristics;model construction and application
《中国烟草学报》 2024 (001)
36-45 / 10
湖北省烟草公司科技项目"基于高光谱成像的中棵烟长势长相及上部烟成熟度判别技术研究"(027Y2021-025)
10.16472/j.chinatobacco.2023.T0014
评论