基于聚类的烟叶标准库和特征选择  

Tobacco Standard Library and Feature Selection Based on the Cluster

作　　者：李航[1] 申金媛[1] 刘润杰[1] 穆晓敏[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450001

出　　处：《黑龙江农业科学》2016年第4期140-143,共4页Heilongjiang Agricultural Sciences

基　　金：河南省烟草公司科技计划资助项目(No.M201335)

摘　　要：为实现烟叶自动化分级,采用聚类算法来剔除烟叶样本中异样样本,通过计算类间的方差,选取方差值大的特征作为有用特征。利用支持向量机模拟人工操作,进行13个等级分部位、分颜色的识别。结果表明:特征选择后的最佳准确率分别为95.52%、98.22%。在提升准确率的同时,减少了输入特征的个数和采集光谱数据所需的时间。There are possibilities that the different samples may exist among the tobacco left samples,so the clustering method is applied to get rid of the different samples.By calculating the variance between the classes,the characteristics of big variance value as a useful feature were selected.Using SVM to simulate the artificial operation,like the recognition of the parts and the colors of thirteen grades of the tobacco left samples.The optima accuracy reached 95.52% and 98.22% respectively after the select of the useful features.Meanwhile,not only the accuracy arising,but the number of input features and the time needed for spectral data were reduced.

关 键 词：烟叶 光谱 聚类分析 特征提取 

分 类 号：S572[农业科学—烟草工业]