基于机器学习算法的烤烟香型分类研究

Study on Classification of Flue-cured Tobacco Based on Machine Learning Methods

为了探索烤烟香型判别分析的方法,采集了我国典型香型的烤烟样本,运用机器学习的方法对训练集和测试集的样本进行了模型拟合,结果表明:对于清香型、浓香型、中间香型拟合最好的机器学习算法为神经网络模型,就香型而言,该模型对于清香型、浓香型评价相对较好,中间香型整体判定效果较低;就数据集而言,在数据准备中分部位进行香型判别较为合理。在具体香型分析判别中,可首选神经网络机器学习算法,为烟草质量评价和卷烟产品研发等提供技术依据。

To explore the methods for flue- cured tobacco types classification,typical tobacco samples were collected from main tobacco-planted areas in China. These samples had been trained and tested by several machine learning methods to model fitting. The results showed that Neural Network algorithm could expressed best prediction from these machine learning methods to predict the flue-cured tobacco types,the Qing and Nong types were trained and tested best by the machine learning method,but the Zhong type was trained and tested worse by such methods,discriminant by vary part tobacco leaves was reasonable in the data preparation. So Neural Network algorithm could be used to predict the flue-cured tobacco types in tobacco quality evaluation and cigarette production.