不同成熟度茄衣烟叶高光谱特征分析及分类模型构建  

Analysis of Hyperspectral Characteristics of Coat-oriented Cigar Tobacco with Different Maturity and Construction of a Classification Model

作　　者：王大彬 孟贵星 王雪松 陈红华 邓建强 向必坤 谭绍安 王瑞 WANG Dabin;MENG Guixing;WANG Xuesong;CHEN Honghua;DENG Jianqiang;XIANG Bikun;TAN Shaoan;WANG Rui(Laboratory of Quality&Safety Risk Assessment for Tobacco,Ministry of Agriculture,Tobacco Research Institute of Chinese Academy of Agricultural Sciences,Qingdao 266101,China;Enshi Tobacco Company of Hubei Province,Enshi,Hubei 445000,China)

机构地区：[1]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 [2]湖北省烟草公司恩施州公司,湖北恩施445000

出　　处：《中国烟草科学》2023年第3期85-91,共7页Chinese Tobacco Science

基　　金：湖北省烟草公司科技项目(027Y2021,2021422800270421)。

摘　　要：为掌握不同成熟度茄衣烟叶的高光谱特征,本研究利用便携式高光谱相机采集3种成熟度[尚熟(SS)、成熟(CS)以及完熟(WS)]茄衣烟叶的高光谱图像并提取光谱数据,通过多元统计分析研究了不同成熟度茄衣烟叶高光谱特征,并构建了支持向量机(SVM)模型用于茄衣烟叶成熟度的分类判别。茄衣烟叶高光谱在400~680 nm、730~1000 nm各自区域内以及400~430 nm与730~1000 nm两个区域之间具有很强的正相关性,在460~730 nm与760~940 nm之间相关性较弱;3个特征值大于1的主成分(PC1~PC3)几乎包含了所有的光谱信息;主成分方差分析结果表明,不同成熟度茄衣烟叶的高光谱反射特征在可见光、红边以及近红外区域均存在统计学差异;SVM对3种成熟度茄衣烟叶的分类准确率达到98%以上,对CS茄衣烟叶的精确率、召回率和F1分数均为100%,对SS和WS茄衣烟叶均达到94%以上。茄衣烟叶高光谱存在多重共线性,具有很好的降维效果,不同成熟度茄衣烟叶的光谱反射特征差异明显,SVM在茄衣烟叶成熟度分类判别方面表现优异,可辅助用于茄衣烟叶适熟采收。To obtain hyperspectral characteristics of different-maturity coat-oriented cigar tobacco,in this study we used a portable hyperspectral camera to collect hyperspectral images of coat-oriented cigar tobacco at three-maturity stages(SS,CS and WS)and extracted their spectral data.The characteristics of hyper spectrum was analyzed by using multivariate statistical analysis,and a supported vector machine(SVM)model was constructed for discriminating the maturity of coat-oriented cigar tobacco.There existed strong correlation for the bands in the region of 400-680 nm,730-1000 nm as well as between 400-430 nm and 730-1000 nm,while the correlation between 460-730 nm and 760-940 nm was weak.Three principal components(PC1,PC2 and PC3)with eigenvalues larger than 1 almost contained all the spectral information,and analysis of variance from principal components indicated that there was significant difference among the three-maturity coat-oriented cigar tobacco in visible light,red edge and near infrared region.For SVM,the accuracy of discriminating cigar maturity was more than 98%,and the precision,recall and F1 scores were 100%for CS cigar as well as more than 94%for SS and WS cigar.The hyper spectrum of cigar had multicollinearity with a good dimension reduction effect.The spectral reflection characteristics of different-maturity coat-oriented cigar tobacco was significantly different.SVM had good performance in discriminating maturity of coat-oriented cigar tobacco,which could be used in ripening harvest of coat-oriented cigar tobacco.

关 键 词：茄衣烟叶 成熟度 高光谱成像 光谱特征 模型构建 

分 类 号：S572.01[农业科学—烟草工业]