基于颜色分布信息的烟叶烟碱含量预测模型评估与比较  

Evaluation and Comparison of Prediction Model for Nicotine Content in Tobacco Leaves Based on Color Distribution Information

作　　者：谢嘉妮 云利军[1,2] 尹晓东 陈载清 XIE Jiani;YUN Lijun;YIN Xiaodong;CHEN Zaiqing(School of Information Science and Technology,Yunnan Normal University,Kunming 650500,China;Computer Vision and Intelligent Control Technology Engineering Research Center of Yunnan Provincial Department of Education,Kunming 650500,China;Yunnan Provincial Tobacco Company,Kunming 650218,China)

机构地区：[1]云南师范大学信息学院,云南昆明650500 [2]云南省教育厅计算机视觉与智能控制技术工程研究中心,云南昆明650500 [3]云南省烟草烟叶公司,云南昆明650218

出　　处：《现代信息科技》2025年第8期132-137,145,共7页Modern Information Technology

基　　金：中国烟草总公司云南省公司科技计划项目(2022530000241026)。

摘　　要：利用光学成像技术对一批已知烟碱含量的烟叶建立图像数据集,并使用神经网络模型U2-Net精准检测烟叶目标,通过提取烟叶目标的颜色分布信息,应用了4种典型的机器学习算法:随机森林(RF)、梯度提升(XGBoost)、多层感知机(MLP)和K-最近邻(KNN),分别对烟叶烟碱含量进行了回归预测。结果显示KNN模型能有效利用颜色分布信息对烟叶烟碱含量进行精准预测,其决定系数R~2值高达97.46%,均方误差(MSE)低至0.020 2,平均绝对误差(MAE)低至0.075 6,证明烟叶颜色分布信息与烟碱含量之间具有显著的相关性,提供了一种有效的烟碱无损检测方法。This paper uses optical imaging technology to establish an image dataset of a batch of tobacco leaves with known nicotine content,and uses the neural network model U2-Net to accurately detect tobacco leaf targets.By extracting the color distribution information of the tobacco leaf targets,four typical Machine Learning algorithms of RF,XGBoost,MLP,and KNN are used to make the regression prediction for the nicotine content of tobacco leaves,respectively.The results indicate that the KNN model can effectively utilize color distribution information to accurately predict the nicotine content of tobacco leaves.The value of determination coefficient R2 is as high as 97.46%,the MSE is as low as 0.0202,and the MAE is as low as 0.0756,indicating a significant correlation between tobacco leaf color distribution information and nicotine content,and providing an effective nondestructive detection method for nicotine.

关 键 词：颜色分布 烟碱预测 回归模型 K-近邻 无损检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP181[自动化与计算机技术—计算机科学与技术]