模型集群分析-随机森林方法在烟叶分类中的应用

Application of Random Forest Method Based on Model Population Analysis in Tobacco Quality Classification

为了解决烟叶外观质量检验和烟叶品质等级评估中主观因素影响过大的问题,首次采用模型集群分析-随机森林方法(MPA-RF)结合近红外光谱建立的烟叶采收成熟度和烤后烟叶等级划分判别模型对烟叶进行了品质分类。结果表明:MPA-RF模型对采收成熟度烟叶样本(数据集A)和不同等级烟叶样本(数据集B)的训练集分类精度分别为96.67%、99.02%,预测模型分类精度分别为100%、96.15%;MPA-RF模型对烟叶的分类准确率明显高于常用的PCA、SVM和RF分类方法。

In order to solve the problem that subjective factors have seriously affected the tobacco appearance quality inspection and tobacco quality grade evaluation, we constructed a classification model which was based on the method of Model Population Analysis-Random Forest （MPA-RF） combined with the near-infrared spectroscopy, and firstly adopted this model to conduct the quality classification of flue-cured tobacco leaves with different maturity grades and different quality grades. The results indicated that MPA-RF model possessed the classification accuracy of 96.67% and 99.02% for the training set of tobacco leaf samples with different maturity grades （data set A） and tobacco leaf samples with different quality grades （data set B）, respectively, and it had the classification accuracy of 100% and 96.15% for the forecasting set of data set A and data set B, respectively. The tobacco leaf quality classification accuracy of MPA-RF model was obviously higher than that of the commonly-used classification methods such as principle component analysis （PCA）, support vector machine （SVM） and random forest （RF）.