基于近红外光谱的醇化雪茄烟叶品种判别模型研究  

Study on Discrimination Model of Alcoholized Cigar Tobacco Varieties Based on Near Infrared Spectroscopy

作　　者：孙利 张毅[1] 孟广云 余彦 高飞 王潞 SUN Li;ZHANG Yi;MENG Guangyun;YU Yan;GAO Fei;WANG Lu(College of Biological Science and Technology,Hunan Agricultural University,Changsha,Hunan 410128,China;Baoshan City Company,Yunnan Tobacco Company,Baoshan,Yunan 678000,China)

机构地区：[1]湖南农业大学生物科学技术学院,湖南长沙4101281 [2]云南省烟草公司保山市公司,云南保山678000

出　　处：《天津农业科学》2024年第4期82-90,共9页Tianjin Agricultural Sciences

基　　金：中国烟草总公司云南省公司科技计划项目(2021530000242040)。

摘　　要：为提高不同醇化后雪茄烟叶品种的判别准确性,采用多元散射校正等预处理算法对光谱数据进行去噪处理,以降低试验、环境和仪器噪音对数据的影响。结合支持向量机、BP神经网络和随机森林建立不同品种的近红外光谱判别模型,通过准确率和混淆矩阵评估模型性能。结果表明:采用SNV+FD预处理算法和CARS特征波长选择算法建立的模型效果最佳,在训练集和预测集上均表现出较高准确性,证实了利用近红外光谱技术快速判别不同醇化后雪茄烟叶品种的可行性。综上,利用近红外光谱技术可实现对不同品种醇化后雪茄烟叶的无损、快速判别,进一步提高雪茄烟叶工业可用性。The aim of this study was to improve the discrimination accuracy of different post-alcoholised cigar tobacco varieties,pre-processing algorithms such as multiple scattering correction were used to denoise the spectral data in order to reduce the influence of experimental,environmental and instrumental noise on the data.Support vector machines,BP neural networks and random forests were combined to establish near-infrared spectral discrimination models for different varieties.The model performance was evaluated by accuracy and confusion matrix.The results showed that the model built with SNV+FD preprocessing algorithm and CARS feature wavelength selection algorithm was the most effective,and showed high accuracy in both training and prediction sets,which confirmed the feasibility of the use of NIR spectroscopy to quickly discriminate different varieties of post-alcoholisation cigar tobacco leaves.In summary,the use of near infrared spectroscopy could realize the non-destructive and rapid discrimination of different varieties of alcoholized cigar tobacco,and further improve the industrial availability of cigar tobacco.

关 键 词：雪茄烟叶 近红外光谱技术 品种判别 支持向量机 随机森林 BP神经网络 

分 类 号：S572[农业科学—烟草工业]