作　　者： (徐东）; (陶欧）; (林兆洲）; (米文娟）; (赵丽莹）; (拱健婷）; (李阳）; (闫永红）;

机构地区： 北京中医药大学中药学院,北京100029

出　　处： 《中华中医药学刊》 2017年第9期2323-2325,共3页

摘　　要： 目的:将机器学习方法引入到电子鼻系统,应用于分类器的优化中,建立一种准确、快捷的中草药智能鉴别手段。方法:以菊科常用八味药材为载体,基于电子鼻建立中药气味特征指纹图谱,联合Best First搜索策略加Cfs Subset Eval评估策略的BC属性筛选法,并结合径向基函数、随机森林等建立分类器。采用十折交叉验证法、外部测试集验证法,展开对智能鉴别系统性能的预测与评估。结果:在传感器数量从12缩减为5的情况下,所建立的分类器仍保持原有正判率(100%),此方案简便、准确;并实现了中草药电子鼻智能鉴别系统的建立。结论:本研究首次呈现了BC属性筛选法对中草药电子鼻智能鉴别系统中分类器的优化;可望为中药智能鉴别提供一种快速、可靠而有效的分析方法。 In the past decades,electronic nose（ E-nose） has been widely used in many areas,such as quality monitoring of agricultural products,on-line detecting of dangerous chemicals,additional agents determination in food consuming,E. T. C. Furthermore,it becomes nowadays more and more popular,that efforts have been put forward to increasing the accuracy of discriminative model combined with different machine learning methods. Among them,article neural network（ ANN） has occupied the important position. Based on these,two ANN methods,namely RBF（ Radial basis function） and RF（ Random forests）,were employed to establish classifiers for signal processing. Besides,there are normally 12 or even more sensors in one E-nose system. In consequence,sensor drift and dimension catastrophe would be the first key issue. In this paper,an intelligent and efficient attributes screening and reducing method were introduced to solve this problem,which is Best First combined with Cfs Subset Eval. As shown in the results,the correct judge rate remained the same as former（ 100%）,although the number of sensor has been decreased from 12 down to 5.Obviously,this new method could not only eliminate the dimension catastrophe so as to quantitatively eliminate interfering signals from sensor drift,but also achieve the identification of different Chinese Material Medica（ CMM） based on their odor fingerprint in E-nose system.

关 键 词： 电子鼻 中草药 分类器 属性筛选