烤烟产地近红外高光谱判别技术  

Near-Infrared Hyperspectral Identification of Flue-Cured Tobacco Producing Areas

作　　者：周岩[1] 江晟 李野[1] 李翔宇 赵鹏[1] ZHOU Yan;JIANG Sheng;LI Ye;LI Xiangyu;ZHAO Peng(School of Physics,Changchun University of Science and Technology,Changchun 130022;Yanbian Prefecture Tobacco Company of Jilin Province,Yanji 133000)

机构地区：[1]长春理工大学物理学院,长春130022 [2]吉林省烟草公司延边州公司,延吉133000

出　　处：《长春理工大学学报（自然科学版）》2023年第4期17-22,共6页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省重点科技研发项目(20180201112GX)。

摘　　要：针对不同品种烤烟分类的问题,提出利用五折交叉验证支持向量机算法对融合可见光-近红外高光谱成像信息进行分析,完成对烤烟的快速鉴别。利用某烟草公司提供的烤烟样本,对多个产区烤烟三个部位进行识别分析。首先利用可见光-近红外高光谱获取多组烤烟光谱特征曲线,通过多元散射校正方法开展预处理优化,建立五折交叉验证支持向量机模型,并采用卡帕系数对样品分类精度进行表征。实验结果表明,基于可见光-近红外高光谱技术的烤烟分类技术可以有效实现对不同品种烤烟的分类鉴别,并且与传统人工方法相比,鉴定结果更加客观和精准。Aiming at the classification of different varieties of flue-cured tobacco,the five-fold cross-validation support vector machine algorithm is proposed to analyze the fused visible-near-infrared hyperspectral imaging information to complete the rapid identification of flue-cured tobacco.Using flue-cured tobacco samples provided by a tobacco company,three parts of flue-cured tobacco in multiple production areas were identified and analyzed.Firstly,multiple sets of flue-cured tobacco spectral characteristic curves were obtained by using visible light-near-infrared hyperspectroscopy,pretreatment optimization was carried out by multivariate scattering correction method,a five-fold cross-verification support vector machine model was established,and the Kappa coefficient was used to characterize the sample classification accuracy.The experimental results show that the flue-cured tobacco classification technology based on visible light-near-infrared hyperspectral technology can effectively realize the classification and identification of different varieties of flue-cured tobacco,and the identification results are more objective and accurate than the traditional manual method.

关 键 词：高光谱 烤烟分类 支持向量机 交叉验证 多元散射校正 

分 类 号：O433.4[机械工程—光学工程]